id
string | submitter
string | authors
string | title
string | comments
string | journal-ref
string | doi
string | report-no
string | categories
string | license
string | abstract
string | versions
list | update_date
timestamp[s] | authors_parsed
sequence |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2412.12280 | Stashu Kozlowski | Stashu Kozlowski, Derek Churchill, Hae-Young Kee | Emergence of quadrupolar order under magnetic field in $5d^2$ double perovskites | 9 pages, 5 figures | Phys. Rev. B 111, 155138 (2025) | null | null | cond-mat.str-el | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Motivated by the time-reversal symmetry breaking signal in muon spin relaxation below a transition temperature without accompanying noticeable magnetic Bragg peaks in $5d^2$ Os double perovskites, a rare ferro-octupolar order was proposed to account for such hidden order. Here we study the phase transitions under a magnetic field in triangular and face-centered cubic lattices using classical Monte Carlo simulations. It is expected that higher-rank moments do not couple linearly to the magnetic field. Consequently, a field applied along the ferro-octupolar order is not anticipated to influence the ordering. However, we observe the emergence of antiferro-quadrupolar ordering mixed with the antiferro-octupolar order (AFQO) due to the field-induced bond-dependent exchange interaction. This field-linear interaction among non-Kramer doublets arises via the coupling to the excited triplet states enabled by the external field. In the triangular lattice, we uncover intriguing vortex states, which could inspire future research into $5d^2$ triangular lattice systems. | [
{
"version": "v1",
"created": "Mon, 16 Dec 2024 19:00:21 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kozlowski",
"Stashu",
""
],
[
"Churchill",
"Derek",
""
],
[
"Kee",
"Hae-Young",
""
]
] |
2412.12633 | Muchen Ju | Muchen Ju, Junjie Ni, Kaixin Wang, Yihan Xiao | Arborescences of Random Covering Graphs | 10 pages,4 figures | null | null | null | math.CO | http://creativecommons.org/licenses/by/4.0/ | A rooted arborescence of a directed graph is a spanning tree directed towards a particular vertex. A recent work of Chepuri et al. showed that the arborescences of a covering graph of a directed graph G are closely related to the arborescences of G. In this paper, we study the weighted sum of arborescences of a random covering graph and give a formula for the expected value, resolving a conjecture of Chepuri et al. | [
{
"version": "v1",
"created": "Tue, 17 Dec 2024 07:56:10 GMT"
},
{
"version": "v2",
"created": "Wed, 18 Dec 2024 10:05:11 GMT"
},
{
"version": "v3",
"created": "Mon, 26 May 2025 18:23:40 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 12:11:35 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ju",
"Muchen",
""
],
[
"Ni",
"Junjie",
""
],
[
"Wang",
"Kaixin",
""
],
[
"Xiao",
"Yihan",
""
]
] |
2412.12974 | Wenhao Sun | Benlei Cui, Wenhao Sun, Xue-Mei Dong, Jingqun Tang, Yi Liu | Attentive Eraser: Unleashing Diffusion Model's Object Removal Potential via Self-Attention Redirection Guidance | Accepted by AAAI 2025(Oral) | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recently, diffusion models have emerged as promising newcomers in the field of generative models, shining brightly in image generation. However, when employed for object removal tasks, they still encounter issues such as generating random artifacts and the incapacity to repaint foreground object areas with appropriate content after remova1l. To tackle these problems, we propose Attentive Eraser, a tuning-free method to empower pre-trained diffusion models for stable and effective object removal. Firstly, in light of the observation that the self-attention maps influence the structure and shape details of the generated images, we propose Attention Activation and Suppression (ASS), which re-engineers the self-attention mechanism within the pre-trained diffusion models based on the given mask, thereby prioritizing the background over the foreground object during the reverse generation process. Moreover, we introduce Self-Attention Redirection Guidance (SARG), which utilizes the self-attention redirected by ASS to guide the generation process, effectively removing foreground objects within the mask while simultaneously generating content that is both plausible and coherent. Experiments demonstrate the stability and effectiveness of Attentive Eraser in object removal across a variety of pre-trained diffusion models, outperforming even training-based methods. Furthermore, Attentive Eraser can be implemented in various diffusion model architectures and checkpoints, enabling excellent scalability. Code is available at https://github.com/Anonym0u3/AttentiveEraser. | [
{
"version": "v1",
"created": "Tue, 17 Dec 2024 14:56:59 GMT"
},
{
"version": "v2",
"created": "Wed, 18 Dec 2024 07:52:14 GMT"
},
{
"version": "v3",
"created": "Thu, 19 Dec 2024 08:41:19 GMT"
},
{
"version": "v4",
"created": "Tue, 11 Mar 2025 07:51:49 GMT"
},
{
"version": "v5",
"created": "Tue, 20 May 2025 04:53:12 GMT"
},
{
"version": "v6",
"created": "Mon, 26 May 2025 14:21:31 GMT"
},
{
"version": "v7",
"created": "Thu, 05 Jun 2025 12:54:00 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Cui",
"Benlei",
""
],
[
"Sun",
"Wenhao",
""
],
[
"Dong",
"Xue-Mei",
""
],
[
"Tang",
"Jingqun",
""
],
[
"Liu",
"Yi",
""
]
] |
2412.14245 | Pavlo Sukhachov | Pavlo Sukhachov, Hans Gl{\o}ckner Giil, Bj{\o}rnulf Brekke, Jacob Linder | Coexistence of $p$-wave Magnetism and Superconductivity | 6+19 pages, 3+11 figures; minor changes + added SM | Phys. Rev. B 111, L220403 (2025) | 10.1103/PhysRevB.111.L220403 | QuSpin 2025 | cond-mat.supr-con | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The symmetry requirements for realizing unconventional compensated magnets with spin-polarized bands such as altermagnets have recently been uncovered. The most recent addition to this family of magnets is parity-odd or $p$-wave magnets. We demonstrate that $p$-wave magnets are perfectly compatible with superconductivity due to the spin polarization of their electron bands and that they induce unexpected spin transport phenomena. We first show that $p$-wave magnetism can coexist with conventional superconductivity regardless of the magnitude of the spin splitting. We then predict that $p$-wave magnets induce a charge-to-spin conversion, which can be strongly enhanced by the presence of superconductivity providing a way to probe the coexistence in experiments. Our results open a new avenue for material combinations with a synergetic relation between spintronics and superconductivity. | [
{
"version": "v1",
"created": "Wed, 18 Dec 2024 19:00:01 GMT"
},
{
"version": "v2",
"created": "Wed, 08 Jan 2025 00:01:36 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Sukhachov",
"Pavlo",
""
],
[
"Giil",
"Hans Gløckner",
""
],
[
"Brekke",
"Bjørnulf",
""
],
[
"Linder",
"Jacob",
""
]
] |
2412.16187 | Minghui Liu | Minghui Liu, Tahseen Rabbani, Tony O'Halloran, Ananth Sankaralingam, Mary-Anne Hartley, Furong Huang, Cornelia Ferm\"uller, Yiannis Aloimonos | HashEvict: A Pre-Attention KV Cache Eviction Strategy using Locality-Sensitive Hashing | 10 pages, 6 figures, 2 tables | null | null | null | cs.LG cs.AI cs.CL cs.DS cs.PF | http://creativecommons.org/licenses/by/4.0/ | Transformer-based large language models (LLMs) use the key-value (KV) cache to significantly accelerate inference by storing the key and value embeddings of past tokens. However, this cache consumes significant GPU memory. In this work, we introduce HashEvict, an algorithm that uses locality-sensitive hashing (LSH) to compress the KV cache. HashEvict quickly locates tokens in the cache that are cosine dissimilar to the current query token. This is achieved by computing the Hamming distance between binarized Gaussian projections of the current token query and cached token keys, with a projection length much smaller than the embedding dimension. We maintain a lightweight binary structure in GPU memory to facilitate these calculations. Unlike existing compression strategies that compute attention to determine token retention, HashEvict makes these decisions pre-attention, thereby reducing computational costs. Additionally, HashEvict is dynamic - at every decoding step, the key and value of the current token replace the embeddings of a token expected to produce the lowest attention score. We demonstrate that HashEvict can compress the KV cache by 30%-70% while maintaining high performance across reasoning, multiple-choice, long-context retrieval and summarization tasks. | [
{
"version": "v1",
"created": "Fri, 13 Dec 2024 06:00:27 GMT"
},
{
"version": "v2",
"created": "Tue, 24 Dec 2024 13:04:45 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 22:37:29 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Liu",
"Minghui",
""
],
[
"Rabbani",
"Tahseen",
""
],
[
"O'Halloran",
"Tony",
""
],
[
"Sankaralingam",
"Ananth",
""
],
[
"Hartley",
"Mary-Anne",
""
],
[
"Huang",
"Furong",
""
],
[
"Fermüller",
"Cornelia",
""
],
[
"Aloimonos",
"Yiannis",
""
]
] |
2412.16387 | Yifeng Fan | Yifeng Fan and Zhizhen Zhao | Information Limits of Joint Community Detection and Finite Group Synchronization | null | null | null | null | cs.IT math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The emerging problem of joint community detection and group synchronization, with applications in signal processing and machine learning, has been extensively studied in recent years. Previous research has predominantly focused on a statistical model that extends the stochastic block model~(SBM) by incorporating additional group transformations. In its simplest form, the model randomly generates a network of size $n$ that consists of two equal-sized communities, where each node $i$ is associated with an unknown group element $g_i^* \in G_M$ for some finite group $G_M$ of order $M$. The connectivity between nodes follows a probability $p$ if they belong to the same community, and a probability $q$ otherwise. Moreover, a group transformation $g_{ij} \in G_M$ is observed on each edge $(i,j)$, where $g_{ij} = g_i^* - g_j^*$ if nodes $i$ and $j$ are within the same community, and $g_{ij} \sim \text{Uniform}(G_M)$ otherwise. The goal of the problem is to recover both the underlying communities and group elements. Under this setting, when $p = a\log n /n$ and $q = b\log n /n $ with $a, b > 0$, we establish the following sharp information-theoretic threshold for exact recovery by maximum likelihood estimation~(MLE): $$ (i):\enspace \frac{a + b}{2} -\sqrt{\frac{ab}{M}} > 1 \quad \text{and} \quad (ii):\enspace a > 2$$ where the exact recovery of communities is possible only if $(i)$ is satisfied, and the recovery of group elements is achieved only if both $(i)$ and $(ii)$ are satisfied. Our theory indicates the recovery of communities greatly benefits from the extra group transformations. Also, it demonstrates a significant performance gap exists between the MLE and all the existing approaches, including algorithms based on semidefinite programming and spectral methods. | [
{
"version": "v1",
"created": "Fri, 20 Dec 2024 22:47:01 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 05:51:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Fan",
"Yifeng",
""
],
[
"Zhao",
"Zhizhen",
""
]
] |
2412.16440 | Pei-Cheng Tung | Pei-Cheng Tung and Ke-Jung Chen | Coevolution of Dwarf Galaxies and Their Circumgalactic Medium Across Cosmic Time | Comments: 20 pages, 17 figures, accepted to ApJ | null | null | null | astro-ph.GA astro-ph.CO | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Dwarf galaxies are thought of as the building blocks of large galaxies such as our Milky Way. This paper presents new high-resolution hydrodynamical simulations of dwarf galaxies and their intergalactic medium with the \texttt{GIZMO} code. Our simulations consider the key physical processes of galaxy evolution, such as gas cooling, chemistry, and stellar and black hole feedback. Unlike the previous work, the initial conditions of our simulations taking the dwarf galaxies of $2-5 \times 10^{10} \, M_\odot$ from the realistic cosmology simulations, \texttt{IllustrisTNG}. We further increase the original resolution of \texttt{IllustrisTNG} by a factor of $\sim 100$ via a particle splitting scheme. Our results show that the evolution of complex multiphase CGM and its metal content is sensitive to the redshift of dwarf galaxies. The accretion of CGM into dwarf galaxies plays a key role in providing $20 \% - 50 \%$ of the star-forming gas and replenishing $40 \% - 70 \%$ of the total mass in the galactic disk. Furthermore, the accretion history of supermassive black holes in the centers of high-$z$ dwarf galaxies shows episodic patterns with high-accreting states close to $\sim 10 \%$ of the Eddington mass accretion rate, implying the rapid growth of supermassive black holes in the early universe, which may be revealed by the coming observations from the James Webb Space Telescope (JWST). | [
{
"version": "v1",
"created": "Sat, 21 Dec 2024 02:04:43 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 03:27:14 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Tung",
"Pei-Cheng",
""
],
[
"Chen",
"Ke-Jung",
""
]
] |
2412.16635 | Daniel Honerkamp | Raphael Schneider, Daniel Honerkamp, Tim Welschehold, Abhinav Valada | Task-Driven Co-Design of Mobile Manipulators | Accepted for publication at RA-L. Project website: https://moma-codesign.cs.uni-freiburg.de/ | null | 10.1109/LRA.2025.3573622 | null | cs.RO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent interest in mobile manipulation has resulted in a wide range of new robot designs. A large family of these designs focuses on modular platforms that combine existing mobile bases with static manipulator arms. They combine these modules by mounting the arm in a tabletop configuration. However, the operating workspaces and heights for common mobile manipulation tasks, such as opening articulated objects, significantly differ from tabletop manipulation tasks. As a result, these standard arm mounting configurations can result in kinematics with restricted joint ranges and motions. To address these problems, we present the first Concurrent Design approach for mobile manipulators to optimize key arm-mounting parameters. Our approach directly targets task performance across representative household tasks by training a powerful multitask-capable reinforcement learning policy in an inner loop while optimizing over a distribution of design configurations guided by Bayesian Optimization and HyperBand (BOHB) in an outer loop. This results in novel designs that significantly improve performance across both seen and unseen test tasks, and outperform designs generated by heuristic-based performance indices that are cheaper to evaluate but only weakly correlated with the motions of interest. We evaluate the physical feasibility of the resulting designs and show that they are practical and remain modular, affordable, and compatible with existing commercial components. We open-source the approach and generated designs to facilitate further improvements of these platforms. | [
{
"version": "v1",
"created": "Sat, 21 Dec 2024 14:00:41 GMT"
},
{
"version": "v2",
"created": "Sat, 17 May 2025 21:14:51 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Schneider",
"Raphael",
""
],
[
"Honerkamp",
"Daniel",
""
],
[
"Welschehold",
"Tim",
""
],
[
"Valada",
"Abhinav",
""
]
] |
2412.18359 | Nick R.D. Zhu | Pujun Liu and Rui-Dong Zhu | Notes on Quasinormal Modes of charged de Sitter Blackholes from Quiver Gauge Theories | 18+13 pages; some further explanations and one table are added in v3 | JHEP 06 (2025) 015 | 10.1007/JHEP06(2025)015 | null | hep-th astro-ph.HE gr-qc hep-ph math-ph math.MP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We give the connection formulae for ordinary differential equations with 5 and 6 (and in principle can be generalized to more) regular singularities from the data of instanton partition functions of quiver gauge theories. We check the consistency of these connection formulae by numerically computing the quasinormal modes (QNMs) of Reissner-Nordstr\"om de Sitter (RN-dS) blackhole. Analytic expressions are obtained for all the families of QNMs, including the photon-sphere modes, dS modes, and near-extremal modes. We also argue that a similar method can be applied to the dS-Kerr-Newman blackhole. | [
{
"version": "v1",
"created": "Tue, 24 Dec 2024 11:43:08 GMT"
},
{
"version": "v2",
"created": "Mon, 13 Jan 2025 02:39:06 GMT"
},
{
"version": "v3",
"created": "Tue, 08 Apr 2025 06:03:42 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Liu",
"Pujun",
""
],
[
"Zhu",
"Rui-Dong",
""
]
] |
2412.18776 | Abyad Enan | Abyad Enan, M Sabbir Salek, Mashrur Chowdhury, Gurcan Comert, Sakib M. Khan, and Reek Majumder | Quantum Annealing-Enhanced Virtual Traffic Lights and its Evaluation Using a Quantum-in-the-Loop Simulation Testbed | This work has been submitted to the ACM Transactions on Internet of Things for possible publication | null | null | null | cs.OH | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Virtual Traffic Light (VTL) is a traffic control method that does not require traffic signal-related infrastructure for roadway intersections. Connected vehicles (CVs) are given right-of-way based on prevailing traffic conditions, such as estimated times of arrival (ETAs) of vehicles, the number of CVs in different approaches, and their emissions. These factors are considered in line with the objectives of the VTL application. Aiming to optimize traffic flow by reducing delays, VTL generates Signal Phase and Timing (SPaT) data for CVs approaching an intersection. Our VTL method considers the delay each CV would cause for other CVs if given the right-of-way. However, the stochastic nature of vehicle arrivals at intersections increases the complexity of the optimization problem, making it challenging for classical computers to determine optimal solutions in real-time. To address this limitation, we develop a VTL method designed to minimize stopped delays for CVs at an intersection by leveraging the efficacies of existing quantum computers that determine the best outcome from all possible combinations. This method employs Quadratic Unconstrained Binary Optimization (QUBO), a mathematical framework commonly used in quantum computing, to formulate the VTL problem as a stopped-delay-minimization challenge. To evaluate our method for roadway traffic with varying traffic volumes, we integrate an open-source microscopic roadway traffic simulator, Simulation for Urban Mobility (SUMO), with a cloud-based D-Wave quantum computer. Our analysis reveals that our quantum computing-supported VTL outperforms the classical optimization-based VTL by significantly reducing stopped delays at intersections and travel time through the roadway sections crossing the intersections. | [
{
"version": "v1",
"created": "Sun, 22 Dec 2024 04:52:46 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 16:23:06 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Enan",
"Abyad",
""
],
[
"Salek",
"M Sabbir",
""
],
[
"Chowdhury",
"Mashrur",
""
],
[
"Comert",
"Gurcan",
""
],
[
"Khan",
"Sakib M.",
""
],
[
"Majumder",
"Reek",
""
]
] |
2412.20113 | Javad Mohamadichamgavi | Jacek Miekisz and Javad Mohamadichamgavi | Intrinsic noise in structured replicator dynamics modelling time delays | 13 pages, 9 figures | Dynamic Games and Applications (2025): 1-14 | 10.1007/s13235-025-00640-8 | null | q-bio.PE | http://creativecommons.org/licenses/by/4.0/ | We construct and analyze structured replicator dynamics of the Snowdrift game. In our model, the offspring is put in juvenile compartments and then mature and join adult compartments with strategy-dependent rates. This is augmented by death rates and hence the population size is bounded. In the corresponding birth-death Markov jump process, rates of leaving juvenile compartments may be interpreted as inverses of averages of exponentially distributed time delays. We observe a novel behavior: for equal average time delays of both strategies, the frequency of cooperators in the quasi-stationary state of a stochastic dynamics is bigger than that in the corresponding stationary state of the deterministic structured replicator dynamics which is actually equal to the critical point of the original replicator equation for the Snowdrift game. In short, time delays favor cooperation in the presence of intrinsic stochastic fluctuations. | [
{
"version": "v1",
"created": "Sat, 28 Dec 2024 10:50:11 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 18:29:02 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Miekisz",
"Jacek",
""
],
[
"Mohamadichamgavi",
"Javad",
""
]
] |
2412.20292 | Mason Kamb | Mason Kamb, Surya Ganguli | An analytic theory of creativity in convolutional diffusion models | null | null | null | null | cs.LG cond-mat.dis-nn cs.AI q-bio.NC stat.ML | http://creativecommons.org/licenses/by/4.0/ | We obtain an analytic, interpretable and predictive theory of creativity in convolutional diffusion models. Indeed, score-matching diffusion models can generate highly original images that lie far from their training data. However, optimal score-matching theory suggests that these models should only be able to produce memorized training examples. To reconcile this theory-experiment gap, we identify two simple inductive biases, locality and equivariance, that: (1) induce a form of combinatorial creativity by preventing optimal score-matching; (2) result in fully analytic, completely mechanistically interpretable, local score (LS) and equivariant local score (ELS) machines that, (3) after calibrating a single time-dependent hyperparameter can quantitatively predict the outputs of trained convolution only diffusion models (like ResNets and UNets) with high accuracy (median $r^2$ of $0.95, 0.94, 0.94, 0.96$ for our top model on CIFAR10, FashionMNIST, MNIST, and CelebA). Our model reveals a locally consistent patch mosaic mechanism of creativity, in which diffusion models create exponentially many novel images by mixing and matching different local training set patches at different scales and image locations. Our theory also partially predicts the outputs of pre-trained self-attention enabled UNets (median $r^2 \sim 0.77$ on CIFAR10), revealing an intriguing role for attention in carving out semantic coherence from local patch mosaics. | [
{
"version": "v1",
"created": "Sat, 28 Dec 2024 22:33:29 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 05:09:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kamb",
"Mason",
""
],
[
"Ganguli",
"Surya",
""
]
] |
2412.20925 | Jiaqi Yang | Jiaqi Yang, Xiaohua Xu, Wei Xie | Active Learning with Variational Quantum Circuits for Quantum Process Tomography | null | null | null | null | quant-ph cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Quantum process tomography (QPT) is a fundamental tool for fully characterizing quantum systems. It relies on querying a set of quantum states as input to the quantum process. Previous QPT methods typically employ a straightforward strategy for randomly selecting quantum states, overlooking differences in informativeness among them. In this work, we propose a general active learning (AL) framework that adaptively selects the most informative subset of quantum states for reconstruction. We design and evaluate various AL algorithms and provide practical guidelines for selecting suitable methods in different scenarios. In particular, we introduce a learning framework that leverages the widely-used variational quantum circuits (VQCs) to perform the QPT task and integrate our AL algorithms into the query step. We demonstrate our algorithms by reconstructing the unitary quantum processes resulting from random quantum circuits with up to seven qubits. Numerical results show that our AL algorithms achieve significantly improved reconstruction, and the improvement increases with the size of the underlying quantum system. Our work opens new avenues for further advancing existing QPT methods. | [
{
"version": "v1",
"created": "Mon, 30 Dec 2024 13:12:56 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 05:45:52 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yang",
"Jiaqi",
""
],
[
"Xu",
"Xiaohua",
""
],
[
"Xie",
"Wei",
""
]
] |
2412.20991 | Andrey Gorbach | Andrey V. Gorbach | Topological gap solitons in equidistant lithium niobate waveguide arrays | Published in the special Issue of Low Temperature Physics dedicated to the 80th anniversary of Prof. Alexander S. Kovalev | null | 10.1063/10.0036751 | null | nlin.PS physics.optics | http://creativecommons.org/licenses/by/4.0/ | Equidistant 1D arrays of thin film lithium niobate waveguides can exhibit non-trivial topology due to a specific interplay between inter- and intra-modal couplings of two families of guided modes. In this work we analyze two-colour spatial solitons, emerging due to $\chi_2$ nonlinear interactions between the modes of non-trivial topology in the fundamental harmonic field, and modes of trivial topology in the second harmonic field. We discuss solitons localized in the bulk of the array (bulk solitons), and at an edge of a finite-size array (edge solitons). The latter emerge due to the nonlinear interactions between a topological edge mode in the fundamental harmonic and bulk modes in the second harmonic. We reveal that for each type of soliton, bulk or edge, there generally exist two families of solutions with different internal structures and ranges of propagation constants. All bulk solitons can only be excited above a certain power threshold dictated by the coupling strength in the second harmonic field and the phase matching between the fundamental and second harmonics. The power threshold for edge solitons generally appears to be much lower, and, by tuning the phase matching, it can be reduced to zero. | [
{
"version": "v1",
"created": "Mon, 30 Dec 2024 14:56:16 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 18:30:09 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Gorbach",
"Andrey V.",
""
]
] |
2501.00092 | David Poland | David Poland, Gordon Rogelberg | Moments and saddles of heavy CFT correlators | 51 pages, 4 figures; V3: Updated with many improvements, clarifications, and corrections | null | null | null | hep-th cond-mat.stat-mech math-ph math.MP | http://creativecommons.org/licenses/by/4.0/ | We study the operator product expansion (OPE) of identical scalars in a conformal four-point correlator as a Stieltjes moment problem, and use Riemann-Liouville type fractional differential operators to generate classical moments from the correlation function. We use crossing symmetry to derive leading and subleading relations between moments in $\Delta$ and $J_2 \equiv \ell(\ell+d-2)$ in the ``heavy" limit of large external scaling dimension, and combine them with constraints from unitarity to derive two-sided bounds on moment sequences in $\Delta$ and the covariance between $\Delta$ and $J_2$. The moment sequences which saturate these bounds produce ``saddle point" solutions to the crossing equations which we identify as particular limits of correlators in a generalized free field (GFF) theory. This motivates us to study perturbations of heavy GFF four-point correlators by way of saddle point analysis, and we show that saddles in the OPE arise from contributions of fixed-length operator families encoded by a decomposition into higher-spin conformal blocks. To apply our techniques, we consider holographic correlators of four identical single scalar fields perturbed by a bulk interaction, and use their first few moments to derive Gaussian weight-interpolating functions that predict the OPE coefficients of interacting double-twist operators in the heavy limit. | [
{
"version": "v1",
"created": "Mon, 30 Dec 2024 19:00:00 GMT"
},
{
"version": "v2",
"created": "Tue, 01 Apr 2025 17:53:15 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 21:05:45 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Poland",
"David",
""
],
[
"Rogelberg",
"Gordon",
""
]
] |
2501.00247 | Min Long | Min Long, Hongyu Lu, Han-Qing Wu, and Zi Yang Meng | Spectra of Magnetoroton and Chiral Graviton Modes of Fractional Chern Insulator | 5+5pages, 4+8 figures | null | null | null | cond-mat.str-el cond-mat.mes-hall quant-ph | http://creativecommons.org/licenses/by/4.0/ | Employing the state-of-the-art time-dependent variational principle (TDVP) algorithm, we compute the spectra of charge-neutral excitations in the $\nu=1/2$ (bosonic) \updated{ and $1/3$ (fermionic) fractional Chern insulator (FCI)} on the Haldane honeycomb lattice model. The magnetoroton visualized from the dynamic density structure factor acquires a minimum gap at finite momentum that can go soft with increasing interaction and give rise to a charge density wave (CDW) at the same wavevector. As the system approaches the FCI-to-CDW transition point, we observe a pronounced sharpening of the roton mode, suggesting that the magnetoroton behaves more like a quasiparticle as it softens. Notably, this occurs while the single-particle gap remains finite. Besides the magnetoroton at finite momentum, we also construct quadrupolar chiral operators in a discrete lattice and resolve the chiral graviton mode around the $\Gamma$ point of the Brillouin zone. Furthermore, we show the different chiralities of the gravitons of FCIs with opposite-sign Hall conductance for the first time. | [
{
"version": "v1",
"created": "Tue, 31 Dec 2024 03:34:20 GMT"
},
{
"version": "v2",
"created": "Wed, 08 Jan 2025 08:29:21 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 01:28:34 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Long",
"Min",
""
],
[
"Lu",
"Hongyu",
""
],
[
"Wu",
"Han-Qing",
""
],
[
"Meng",
"Zi Yang",
""
]
] |
2501.00286 | Hiroshi Suzuki | Motokazu Abe, Okuto Morikawa, and Hiroshi Suzuki | Monte Carlo Simulation of the $SU(2)/\mathbb{Z}_2$ Yang--Mills Theory | 24 pages, many figures. The final version to appear in PTEP | null | null | KYUSHU-HET-307, RIKEN-iTHEMS-Report-25 | hep-lat hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We carry out a hybrid Monte Carlo (HMC) simulation of the $SU(2)/\mathbb{Z}_2$ Yang--Mills theory in which the $\mathbb{Z}_N$ 2-form flat gauge field (the 't~Hooft flux) is explicitly treated as one of the dynamical variables. We observe that our HMC algorithm in the $SU(2)/\mathbb{Z}_2$ theory drastically reduces autocorrelation lengths of the topological charge and of a physical quantity which couples to slow modes in the conventional HMC simulation of the $SU(2)$ theory. Provided that sufficiently large lattice volumes are available, therefore, the HMC algorithm of the $SU(N)/\mathbb{Z}_N$ theory could be employed as an alternative for the simulation of the $SU(N)$ Yang--Mills theory, because local observables are expected to be insensitive to the difference between $SU(N)$ and~$SU(N)/\mathbb{Z}_N$ in the large volume limit. A possible method to incorporate quarks [fermions in the fundamental representation of~$SU(N)$ with the baryon number~$1/N$] in this framework is also considered. | [
{
"version": "v1",
"created": "Tue, 31 Dec 2024 05:37:25 GMT"
},
{
"version": "v2",
"created": "Thu, 09 Jan 2025 09:27:26 GMT"
},
{
"version": "v3",
"created": "Sun, 20 Apr 2025 04:35:14 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 09:29:29 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Abe",
"Motokazu",
""
],
[
"Morikawa",
"Okuto",
""
],
[
"Suzuki",
"Hiroshi",
""
]
] |
2501.00321 | Ling Fu | Ling Fu, Zhebin Kuang, Jiajun Song, Mingxin Huang, Biao Yang, Yuzhe Li, Linghao Zhu, Qidi Luo, Xinyu Wang, Hao Lu, Zhang Li, Guozhi Tang, Bin Shan, Chunhui Lin, Qi Liu, Binghong Wu, Hao Feng, Hao Liu, Can Huang, Jingqun Tang, Wei Chen, Lianwen Jin, Yuliang Liu, and Xiang Bai | OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning | null | null | null | null | cs.CV cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Scoring the Optical Character Recognition (OCR) capabilities of Large Multimodal Models (LMMs) has witnessed growing interest. Existing benchmarks have highlighted the impressive performance of LMMs in text recognition; however, their abilities in certain challenging tasks, such as text localization, handwritten content extraction, and logical reasoning, remain underexplored. To bridge this gap, we introduce OCRBench v2, a large-scale bilingual text-centric benchmark with currently the most comprehensive set of tasks (4x more tasks than the previous multi-scene benchmark OCRBench), the widest coverage of scenarios (31 diverse scenarios), and thorough evaluation metrics, with 10,000 human-verified question-answering pairs and a high proportion of difficult samples. Moreover, we construct a private test set with 1,500 manually annotated images. The consistent evaluation trends observed across both public and private test sets validate the OCRBench v2's reliability. After carefully benchmarking state-of-the-art LMMs, we find that most LMMs score below 50 (100 in total) and suffer from five-type limitations, including less frequently encountered text recognition, fine-grained perception, layout perception, complex element parsing, and logical reasoning. The project website is at: https://99franklin.github.io/ocrbench_v2/ | [
{
"version": "v1",
"created": "Tue, 31 Dec 2024 07:32:35 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 02:59:05 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Fu",
"Ling",
""
],
[
"Kuang",
"Zhebin",
""
],
[
"Song",
"Jiajun",
""
],
[
"Huang",
"Mingxin",
""
],
[
"Yang",
"Biao",
""
],
[
"Li",
"Yuzhe",
""
],
[
"Zhu",
"Linghao",
""
],
[
"Luo",
"Qidi",
""
],
[
"Wang",
"Xinyu",
""
],
[
"Lu",
"Hao",
""
],
[
"Li",
"Zhang",
""
],
[
"Tang",
"Guozhi",
""
],
[
"Shan",
"Bin",
""
],
[
"Lin",
"Chunhui",
""
],
[
"Liu",
"Qi",
""
],
[
"Wu",
"Binghong",
""
],
[
"Feng",
"Hao",
""
],
[
"Liu",
"Hao",
""
],
[
"Huang",
"Can",
""
],
[
"Tang",
"Jingqun",
""
],
[
"Chen",
"Wei",
""
],
[
"Jin",
"Lianwen",
""
],
[
"Liu",
"Yuliang",
""
],
[
"Bai",
"Xiang",
""
]
] |
2501.00325 | Michal \v{S}indler | Michal \v{S}indler and Franti\v{s}ek Herman and Filip Kadlec and Christelle Kadlec | Deeper insight into the terahertz response of conventional superconductors under magnetic field | 7 pages, 6 figures, This revision incorporates changes made in response to referee comments for Physical Review B | null | null | null | cond-mat.supr-con | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We investigate the terahertz conductivity of conventional superconductors in Voigt and Faraday magneto-optical configurations. First, we review theoretical approaches describing the fundamental processes of suppression of superconductivity in magnetic field and how the in-gap states are filled. In the Voigt geometry, thin superconducting films are fully penetrated by the magnetic field which interacts with the spin, thus modifying the magnitudes of the optical gap and of the density of the condensate. In this configuration, we provide an alternative description of the recent experiments showing the gapless conductivity of a Nb film measured by Lee ${\it et\,al.}$ [Nat. Commun. 14,2737 (2023)], which better fits their data for magnetic fields above 1 T. In the Faraday geometry, we measured and analyzed the terahertz conductivity of three NbN films with varying thicknesses using the Maxwell-Garnett model, treating vortices as normal-state inclusions within a superconducting matrix. In both geometries, the optical conductivity can be comprehensively described by the model of Herman and Hlubina [Phys. Rev. B 96, 014509 (2017)] involving pair-conserving, and magnetic-field-dependent pair-breaking disorder scattering processes. | [
{
"version": "v1",
"created": "Tue, 31 Dec 2024 07:49:09 GMT"
},
{
"version": "v2",
"created": "Fri, 24 Jan 2025 17:51:39 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 17:23:58 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Šindler",
"Michal",
""
],
[
"Herman",
"František",
""
],
[
"Kadlec",
"Filip",
""
],
[
"Kadlec",
"Christelle",
""
]
] |
2501.00961 | Chenyu You | Chenyu You, Haocheng Dai, Yifei Min, Jasjeet S. Sekhon, Sarang Joshi, James S. Duncan | Uncovering Memorization Effect in the Presence of Spurious Correlations | Accepted by Nature Communications | null | null | null | cs.LG cs.AI cs.CV eess.IV | http://creativecommons.org/licenses/by/4.0/ | Machine learning models often rely on simple spurious features -- patterns in training data that correlate with targets but are not causally related to them, like image backgrounds in foreground classification. This reliance typically leads to imbalanced test performance across minority and majority groups. In this work, we take a closer look at the fundamental cause of such imbalanced performance through the lens of memorization, which refers to the ability to predict accurately on atypical examples (minority groups) in the training set but failing in achieving the same accuracy in the testing set. This paper systematically shows the ubiquitous existence of spurious features in a small set of neurons within the network, providing the first-ever evidence that memorization may contribute to imbalanced group performance. Through three experimental sources of converging empirical evidence, we find the property of a small subset of neurons or channels in memorizing minority group information. Inspired by these findings, we hypothesize that spurious memorization, concentrated within a small subset of neurons, plays a key role in driving imbalanced group performance. To further substantiate this hypothesis, we show that eliminating these unnecessary spurious memorization patterns via a novel framework during training can significantly affect the model performance on minority groups. Our experimental results across various architectures and benchmarks offer new insights on how neural networks encode core and spurious knowledge, laying the groundwork for future research in demystifying robustness to spurious correlation. | [
{
"version": "v1",
"created": "Wed, 01 Jan 2025 21:45:00 GMT"
},
{
"version": "v2",
"created": "Wed, 15 Jan 2025 06:46:51 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 00:06:59 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"You",
"Chenyu",
""
],
[
"Dai",
"Haocheng",
""
],
[
"Min",
"Yifei",
""
],
[
"Sekhon",
"Jasjeet S.",
""
],
[
"Joshi",
"Sarang",
""
],
[
"Duncan",
"James S.",
""
]
] |
2501.02556 | Yi Zhong | Yi Zhong, Xiaohang Zhou, Ke Feng | Spatial Network Calculus: Toward Deterministic Wireless Networking | 13 pages, 8 figures | null | null | null | cs.NI cs.IT math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper extends the classical network calculus to spatial scenarios, focusing on wireless networks with differentiated services and varying transmit power levels. Building on a spatial network calculus, a prior extension of network calculus to spatial settings, we propose a generalized framework by introducing regulations for stationary marked point processes. The regulations correspond to two key constraints: the total transmit power of all transmitters within a spatial region and the cumulative received power at a receiver, which we refer to as ball regulation and shot-noise regulation, respectively. Then we prove the equivalence of ball regulation and shot-noise regulation for stationary marked point processes and establish a universal lower bound on the performance of all network links under these constraints. This framework is applicable to diverse network scenarios, as demonstrated by the analysis of performance guarantees for networks with multi-class users. In addition, we propose an SINR-based power control scheme adapted to user traffic, which ensures differentiated quality of service (QoS) for different user classes. | [
{
"version": "v1",
"created": "Sun, 05 Jan 2025 14:16:01 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Feb 2025 16:19:33 GMT"
},
{
"version": "v3",
"created": "Thu, 06 Feb 2025 09:39:53 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 13:04:33 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhong",
"Yi",
""
],
[
"Zhou",
"Xiaohang",
""
],
[
"Feng",
"Ke",
""
]
] |
2501.02785 | Shweta Agarwala | Sugandha Saxena, S. N. Prasad, Ashwin M Polnaya, Shweta Agarwala | Hybrid deep convolution model for lung cancer detection with transfer learning | Authors realized mistake in the model. Also some data was misinterpreted | null | null | null | cs.CV cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Advances in healthcare research have significantly enhanced our understanding of disease mechanisms, diagnostic precision, and therapeutic options. Yet, lung cancer remains one of the leading causes of cancer-related mortality worldwide due to challenges in early and accurate diagnosis. While current lung cancer detection models show promise, there is considerable potential for further improving the accuracy for timely intervention. To address this challenge, we introduce a hybrid deep convolution model leveraging transfer learning, named the Maximum Sensitivity Neural Network (MSNN). MSNN is designed to improve the precision of lung cancer detection by refining sensitivity and specificity. This model has surpassed existing deep learning approaches through experimental validation, achieving an accuracy of 98% and a sensitivity of 97%. By overlaying sensitivity maps onto lung Computed Tomography (CT) scans, it enables the visualization of regions most indicative of malignant or benign classifications. This innovative method demonstrates exceptional performance in distinguishing lung cancer with minimal false positives, thereby enhancing the accuracy of medical diagnoses. | [
{
"version": "v1",
"created": "Mon, 06 Jan 2025 06:01:01 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 10:06:50 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Saxena",
"Sugandha",
""
],
[
"Prasad",
"S. N.",
""
],
[
"Polnaya",
"Ashwin M",
""
],
[
"Agarwala",
"Shweta",
""
]
] |
2501.03329 | Mandev Bhullar | Mandev Bhullar, Haoting Xu, Hae-Young Kee | Field-Induced Ordered Phases in Anisotropic Spin-1/2 Kitaev Chains | 15 pages, 10 figures | Phys. Rev. B 111, 104439 (2025) | null | null | cond-mat.str-el | http://creativecommons.org/licenses/by/4.0/ | Motivated by intense research on two-dimensional spin-1/2 Kitaev materials, Kitaev spin chains and ladders, though geometrically limited, have been studied for their numerical simplicity and insights into extended Kitaev models. The phase diagrams under the magnetic field were also explored for these quasi-one dimensional models. For an isotropic Kitaev chain, it was found that a magnetic field polarizes the ground state except along the symmetric field angle, where the chain is found to remain gapless up to a critical field strength where it enters an intriguing soliton phase before reaching the polarized state at higher field strengths. Here we study an anisotropic Kitaev chain under a magnetic field using the density matrix renormalization group technique, where the ground state has a macroscopic degeneracy with a finite gap in the absence of the magnetic field. When the field is mainly aligned parallel to the strong bond, four-site and large unit-cell ordered phases arise. In a certain angle of the field, another ordered phase characterized by a uniform chirality with six-site periodicity emerges. We employ a perturbation theory to understand such field-induced ordered phases. The effective model uncovers the presence of transverse Ising and Dzyaloshinskii-Moriya interactions between unit cells, as well as further-neighbor Ising interaction induced by the magnetic field, which collectively explain the mechanisms behind these ordered states. Open questions and challenges are also discussed. | [
{
"version": "v1",
"created": "Mon, 06 Jan 2025 19:00:21 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Bhullar",
"Mandev",
""
],
[
"Xu",
"Haoting",
""
],
[
"Kee",
"Hae-Young",
""
]
] |
2501.03470 | Jie Wang | Feng Guo and Jie Wang | Positivstellens\"atze for polynomial matrices with universal quantifiers | 28 pages, 2 tables | null | null | null | math.OC | http://creativecommons.org/licenses/by/4.0/ | This paper investigates Positivstellens\"atze for polynomial matrices subject to universally quantified polynomial matrix inequality constraints. We first establish a matrix-valued Positivstellensatz under the Archimedean condition, incorporating universal quantifiers. For scalar-valued polynomial objectives, we further develop a sparse Positivstellensatz that leverages correlative sparsity patterns within these quantified constraints. Moving beyond the Archimedean framework, we then derive a series of generalized Positivstellens\"atze under analogous settings. These results collectively unify and extend foundational theorems in three distinct contexts: classical polynomial Positivstellens\"atze, their universally quantified counterparts, and matrix polynomial formulations. Applications of the established Positivstellens\"atze to robust polynomial matrix optimization are also discussed. | [
{
"version": "v1",
"created": "Tue, 07 Jan 2025 02:12:00 GMT"
},
{
"version": "v2",
"created": "Fri, 31 Jan 2025 14:25:23 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 15:50:50 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Guo",
"Feng",
""
],
[
"Wang",
"Jie",
""
]
] |
2501.03650 | Jan-Christoph Deinert | Gulloo Lal Prajapati, Sergey Kovalev, Igor Ilyakov, Atiqa Arshad, Gaurav Dubey, Ketan S. Navale, Dhanvir Singh Rana, Jan-Christoph Deinert | Terahertz-driven ultrafast dynamics of rare-earth nickelates by controlling only the charge degree of freedom | null | null | 10.1002/adfm.202425867 | null | cond-mat.str-el cond-mat.mtrl-sci | http://creativecommons.org/licenses/by/4.0/ | An important strategy for understanding the microscopic physics of strongly correlated systems and enhancing their technological potential is to selectively drive the fundamental degrees of freedom out of equilibrium. Intense terahertz (THz) pulses with photon energies of a few meV, can not only serve this purpose but also unravel their electronic and quantum nature. Here, we present THz-driven ultrafast dynamics of rare-earth nickelates $\text{RNiO}_{\text{3}}$, $\text{R}$ = rare-earth atom) - a prototype system to study the Mott insulator-metal transition (IMT). The THz drive of its Mott insulating state induces instantaneous IMT via quantum tunneling of valence electrons across the bandgap while the THz drive of its correlated metallic state leads to overall heating of the conduction electrons. The subsequent relaxations of excited electrons in these two states occur via a two-step process (electron-phonon thermalization and recovery of the charge-ordered insulating state) and a one-step process (electron-phonon scattering), respectively. The relaxation dynamics of the electrons and the absence of acoustic phonon modes, in particular, suggest that the THz photons drive only the charge degree of freedom. The purely electronic, ultrafast and local nature of the THz-induced IMT offers its applications in opto-electronics with enhanced performance and minimal device size and heat dissipation. | [
{
"version": "v1",
"created": "Tue, 07 Jan 2025 09:32:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Prajapati",
"Gulloo Lal",
""
],
[
"Kovalev",
"Sergey",
""
],
[
"Ilyakov",
"Igor",
""
],
[
"Arshad",
"Atiqa",
""
],
[
"Dubey",
"Gaurav",
""
],
[
"Navale",
"Ketan S.",
""
],
[
"Rana",
"Dhanvir Singh",
""
],
[
"Deinert",
"Jan-Christoph",
""
]
] |
2501.05460 | Gursimran Singh | Gursimran Singh, Xinglu Wang, Yifan Hu, Timothy Yu, Linzi Xing, Wei Jiang, Zhefeng Wang, Xiaolong Bai, Yi Li, Ying Xiong, Yong Zhang, Zhenan Fan | Efficiently Serving Large Multimodal Models Using EPD Disaggregation | 17 pages, 12 figures, 9 tables | International Conference on Machine Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025 | null | null | cs.DC cs.AI cs.CV cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Multimodal Models (LMMs) extend Large Language Models (LLMs) by handling diverse inputs such as images, audio, and video, but at the cost of adding a multimodal encoding stage that increases both computational and memory overhead. This step negatively affects key Service Level Objectives (SLOs), such as time to first token (TTFT) and time per output token (TPOT). We introduce Encode-Prefill-Decode (EPD) Disaggregation, a novel framework that separates the encoding, prefill, and decode stages onto dedicated resources. Unlike current systems, which bundle encoding and prefill together, our approach decouples these steps, unlocking new opportunities and optimizations. These include a mechanism to cache multimedia tokens for efficient transfer, a novel way to parallelize the encoding load within a request, a module for optimal resource allocation for disaggregated serving, and a novel role-switching method to handle changing workload characteristics. Experimental evaluations with popular LMMs show substantial gains in memory efficiency (up to 15x lower peak memory utilization), batch sizes (up to 22x larger), 10x more images per request, and 2.2x larger KV caches. Furthermore, it leads to significant improvements in SLO attainment (up to 90-100% improvement) and TTFT (up to 71% reduction), compared to systems that do not disaggregate. The code is available at https://github.com/vbdi/epdserve. | [
{
"version": "v1",
"created": "Wed, 25 Dec 2024 10:11:31 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Feb 2025 22:55:47 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 04:21:30 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Singh",
"Gursimran",
""
],
[
"Wang",
"Xinglu",
""
],
[
"Hu",
"Yifan",
""
],
[
"Yu",
"Timothy",
""
],
[
"Xing",
"Linzi",
""
],
[
"Jiang",
"Wei",
""
],
[
"Wang",
"Zhefeng",
""
],
[
"Bai",
"Xiaolong",
""
],
[
"Li",
"Yi",
""
],
[
"Xiong",
"Ying",
""
],
[
"Zhang",
"Yong",
""
],
[
"Fan",
"Zhenan",
""
]
] |
2501.05601 | Adrian Marius Dumitran | Adrian Marius Dumitran, Adrian-Catalin Badea, Stefan-Gabriel Muscalu, Angela-Liliana Dumitran, Stefan-Cosmin Dascalescu, Radu-Sebastian Amarie | Exploring Large Language Models for Translating Romanian Computational Problems into English | 12 pages | null | 10.1109/INISTA62901.2024.10683837 | IEEE Xplore: 24 September 2024 | cs.CL | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Recent studies have suggested that large language models (LLMs) underperform on mathematical and computer science tasks when these problems are translated from Romanian into English, compared to their original Romanian format. Accurate translation is critical for applications ranging from automatic translations in programming competitions to the creation of high-quality educational materials, as well as minimizing errors or fraud in human translations. This study shows that robust large language models (LLMs) can maintain or even enhance their performance in translating less common languages when given well-structured prompts. Our findings suggest that LLMs, with appropriate supervision, can be reliably used for the automatic translation of IOI (International Olympiad in Informatics)-style tasks. We evaluate several translation methods across multiple LLMs, including OpenRoLLM, Llama 3.1 8B, Llama 3.2 3B and GPT-4o, assessing their translation accuracy and performance stability through repeated runs. Additionally, we augment the OJI (Romanian County-Level Informatics Olympiad) Romanian dataset with accurate English translations, enhancing its utility for future LLM training and evaluation. Through detailed syntactic and semantic analyses, we confirm that with human oversight, LLMs can serve as a viable solution for multilingual problem-solving. We also compare the translation quality of LLMs against human translators, as evaluated by a certified expert, underscoring the potential of LLMs in realworld scenarios. | [
{
"version": "v1",
"created": "Thu, 09 Jan 2025 22:17:44 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Dumitran",
"Adrian Marius",
""
],
[
"Badea",
"Adrian-Catalin",
""
],
[
"Muscalu",
"Stefan-Gabriel",
""
],
[
"Dumitran",
"Angela-Liliana",
""
],
[
"Dascalescu",
"Stefan-Cosmin",
""
],
[
"Amarie",
"Radu-Sebastian",
""
]
] |
2501.07182 | Rozin Hasin | Rozin Hasin | Unveiling Voices: A Co-Hashtag Analysis of TikTok Discourse on the 2023 Israel-Palestine Crisis | null | null | null | null | cs.SI cs.CY cs.HC cs.IT cs.MM math.IT | http://creativecommons.org/licenses/by-nc-nd/4.0/ | TikTok has gradually become one of the most pervasive social media platforms in our daily lives. While much can be said about the merits of platforms such as TikTok, there is a different kind of attention paid towards the political affect of social media today compared to its impact on other aspects of modern networked reality. I explored how users on TikTok discussed the crisis in Palestine that worsened in 2023. Using network analysis, I situate keywords representing the conflict and categorize them thematically based on a coding schema derived from politically and ideologically differentiable stances. I conclude that activism and propaganda are contending amongst themselves in the thriving space afforded by TikTok today. | [
{
"version": "v1",
"created": "Mon, 13 Jan 2025 10:25:58 GMT"
},
{
"version": "v2",
"created": "Sat, 22 Feb 2025 12:10:10 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Hasin",
"Rozin",
""
]
] |
2501.07301 | Zhenru Zhang | Zhenru Zhang, Chujie Zheng, Yangzhen Wu, Beichen Zhang, Runji Lin, Bowen Yu, Dayiheng Liu, Jingren Zhou, Junyang Lin | The Lessons of Developing Process Reward Models in Mathematical Reasoning | null | null | null | null | cs.CL cs.AI cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Process Reward Models (PRMs) emerge as a promising approach for process supervision in mathematical reasoning of Large Language Models (LLMs), which aim to identify and mitigate intermediate errors in the reasoning processes. However, the development of effective PRMs faces significant challenges, particularly in data annotation and evaluation methodologies. In this paper, through extensive experiments, we demonstrate that commonly used Monte Carlo (MC) estimation-based data synthesis for PRMs typically yields inferior performance and generalization compared to LLM-as-a-judge and human annotation methods. MC estimation relies on completion models to evaluate current-step correctness, leading to inaccurate step verification. Furthermore, we identify potential biases in conventional Best-of-N (BoN) evaluation strategies for PRMs: (1) The unreliable policy models generate responses with correct answers but flawed processes, leading to a misalignment between the evaluation criteria of BoN and the PRM objectives of process verification. (2) The tolerance of PRMs of such responses leads to inflated BoN scores. (3) Existing PRMs have a significant proportion of minimum scores concentrated on the final answer steps, revealing the shift from process to outcome-based assessment in BoN Optimized PRMs. To address these challenges, we develop a consensus filtering mechanism that effectively integrates MC estimation with LLM-as-a-judge and advocates a more comprehensive evaluation framework that combines response-level and step-level metrics. Based on the mechanisms, we significantly improve both model performance and data efficiency in the BoN evaluation and the step-wise error identification task. Finally, we release a new state-of-the-art PRM that outperforms existing open-source alternatives and provides practical guidelines for future research in building process supervision models. | [
{
"version": "v1",
"created": "Mon, 13 Jan 2025 13:10:16 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 16:34:24 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Zhenru",
""
],
[
"Zheng",
"Chujie",
""
],
[
"Wu",
"Yangzhen",
""
],
[
"Zhang",
"Beichen",
""
],
[
"Lin",
"Runji",
""
],
[
"Yu",
"Bowen",
""
],
[
"Liu",
"Dayiheng",
""
],
[
"Zhou",
"Jingren",
""
],
[
"Lin",
"Junyang",
""
]
] |
2501.07327 | Nicol\`o Alessandro Girardini | Nicol\`o Alessandro Girardini, Antonio Longa, Gaia Trebucchi, Giulia Cencetti, Andrea Passerini, Bruno Lepri | Community Aware Temporal Network Generation | null | null | null | null | cs.SI physics.soc-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The advantages of temporal networks in capturing complex dynamics, such as diffusion and contagion, has led to breakthroughs in real world systems across numerous fields. In the case of human behavior, face-to-face interaction networks enable us to understand the dynamics of how communities emerge and evolve in time through the interactions, which is crucial in fields like epidemics, sociological studies and urban science. However, state-of-the-art datasets suffer from a number of drawbacks, such as short time-span for data collection and a small number of participants. Moreover, concerns arise for the participants' privacy and the data collection costs. Over the past years, many successful algorithms for static networks generation have been proposed, but they often do not tackle the social structure of interactions or their temporal aspect. In this work, we extend a recent network generation approach to capture the evolution of interactions between different communities. Our method labels nodes based on their community affiliation and constructs surrogate networks that reflect the interactions of the original temporal networks between nodes with different labels. This enables the generation of synthetic networks that replicate realistic behaviors. We validate our approach by comparing structural measures between the original and generated networks across multiple face-to-face interaction datasets. | [
{
"version": "v1",
"created": "Mon, 13 Jan 2025 13:41:33 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:12:05 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Girardini",
"Nicolò Alessandro",
""
],
[
"Longa",
"Antonio",
""
],
[
"Trebucchi",
"Gaia",
""
],
[
"Cencetti",
"Giulia",
""
],
[
"Passerini",
"Andrea",
""
],
[
"Lepri",
"Bruno",
""
]
] |
2501.07514 | Tinghan Zhang | Tinghan Zhang | Estimating Sequential Search Models Based on a Partial Ranking Representation | null | null | null | null | econ.EM | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The rapid growth of online shopping has made consumer search data increasingly available, opening up new possibilities for empirical research. Sequential search models offer a structured approach for analyzing such data, but their estimation remains difficult. This is because consumers make optimal decisions based on private information revealed in search, which is not observed in typical data. As a result, the model's likelihood function involves high-dimensional integrals that require intensive simulation. This paper introduces a new representation that shows a consumer's optimal search decision-making can be recast as a partial ranking over all actions available throughout the consumer's search process. This reformulation yields the same choice probabilities as the original model but leads to a simpler likelihood function that relies less on simulation. Based on this insight, we provide identification arguments and propose a modified GHK-style simulator that improves both estimation performances and ease of implementation. The proposed approach also generalizes to a wide range of model variants, including those with incomplete search data and structural extensions such as search with product discovery. It enables a tractable and unified estimation strategy across different settings in sequential search models, offering both a new perspective on understanding sequential search and a practical tool for its application. | [
{
"version": "v1",
"created": "Mon, 13 Jan 2025 17:31:42 GMT"
},
{
"version": "v2",
"created": "Wed, 15 Jan 2025 20:05:20 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 16:17:16 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Tinghan",
""
]
] |
2501.07973 | Attila R\'epai | Attila R\'epai (1), S\'andor F\"oldi (2), P\'eter S\'otonyi (3), Gy\"orgy Cserey (2) ((1) Jedlik Innovation Ltd., Budapest, Hungary, (2) Faculty of Information Technology and Bionics, P\'azm\'any P\'eter Catholic University, Budapest, Hungary, (3) Faculty of Medicine, Department of Vascular and Endovascular Surgery, Semmelweis University, Budapest, Hungary) | An Open Source Validation System for Continuous Arterial Blood Pressure Measuring Sensors | 8 pages, 5 figures. For associated repositories see https://github.com/repat8/cam-bpw-sim-hardware and https://github.com/repat8/cam-bpw-sim | null | null | null | physics.med-ph cs.SY eess.SY | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Measuring the blood pressure waveform is becoming a more frequently studied area. The development of sensor technologies opens many new ways to be able to measure high-quality signals. The development of such an aim-specific sensor can be time-consuming, expensive, and difficult to test or validate with known and consistent waveforms. In this paper, we present an open source blood pressure waveform simulator with an open source Python validation package to reduce development costs for early-stage sensor development and research. The simulator mainly consists of 3D printed parts which technology has become a widely available and cheap solution. The core part of the simulator is a 3D printed cam that can be generated based on real blood pressure waveforms. The validation framework can create a detailed comparison between the signal waveform used to design the cam and the measured time series from the sensor being validated. The presented simulator proved to be robust and accurate in short- and long-term use, as it produced the signal waveform consistently and accurately. To validate this solution, a 3D force sensor was used, which was proven earlier to be able to measure high-quality blood pressure waveforms on the radial artery at the wrist. The results showed high similarity between the measured and the nominal waveforms, meaning that comparing the normalized signals, the RMSE value ranged from $0.0276 \pm 0.0047$ to $0.0212 \pm 0.0023$, and the Pearson correlation ranged from $0.9933 \pm 0.0027$ to $0.9978 \pm 0.0005$. Our validation framework is available at https://github.com/repat8/cam-bpw-sim. Our hardware framework, which allows reproduction of the presented solution, is available at https://github.com/repat8/cam-bpw-sim-hardware. The entire design is an open source project and was developed using free software. | [
{
"version": "v1",
"created": "Tue, 14 Jan 2025 09:46:18 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 11:47:59 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Répai",
"Attila",
"",
"Jedlik Innovation Ltd., Budapest, Hungary"
],
[
"Földi",
"Sándor",
"",
"Faculty of Information Technology and Bionics, Pázmány Péter Catholic University, Budapest, Hungary"
],
[
"Sótonyi",
"Péter",
"",
"Faculty of Medicine, Department of Vascular and Endovascular Surgery, Semmelweis University, Budapest, Hungary"
],
[
"Cserey",
"György",
"",
"Faculty of Information Technology and Bionics, Pázmány Péter Catholic University, Budapest, Hungary"
]
] |
2501.08230 | Bruno Valeixo Bento | Bruno Valeixo Bento, Jo\~ao F. Melo | EFT & Species Scale: Friends or foes? | 43 pages, 1 appendix; v2 added references and corrected discussion on convergence of KK towers | J. High Energ. Phys. 2025, 212 (2025) | 10.1007/JHEP05(2025)212 | null | hep-th | http://creativecommons.org/licenses/by/4.0/ | Recently the notion that quantum gravity effects could manifest at scales much lower than the Planck scale has seen an intense Swamplandish revival. Dozens of works have explored how the so-called species scale -- at which an effective description of gravity must break down -- relates to String Theory and the Swampland conjectures. In particular, the interplay between this scale and the abundant towers of states becoming lighter in asymptotic regions of moduli spaces has proved to be key in understanding the real scale of quantum gravity. Nevertheless concerns have been raised regarding the validity of using infinite towers of states when estimating this scale within Effective Field Theory and, more precisely, the consistency of cutting the tower part way through in a framework that relies on a clear separation of scales. In this work we take an EFT point-of-view and provide a detailed perturbative derivation of the species scale -- by computing the 1-loop graviton propagator in the presence of many fields -- thereby clarifying common sources of confusion in the literature. Not only do we clarify the setup, assumptions and regimes of validity of the result, but more importantly apply the same methods to an infinite tower of states. We show how each state in the tower contributes to the species scale and how the procedure of counting only ''light fields'' can be compatible with not cutting the tower, thereby maintaining the harmony between infinite towers and EFTs even in the context of the species scale. | [
{
"version": "v1",
"created": "Tue, 14 Jan 2025 16:16:06 GMT"
},
{
"version": "v2",
"created": "Tue, 28 Jan 2025 14:35:22 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Bento",
"Bruno Valeixo",
""
],
[
"Melo",
"João F.",
""
]
] |
2501.08301 | Nick Tusay | Nick Tusay, Jason T. Wright, Thomas G. Beatty, Steve Desch, Knicole Col\'on, Tushar Mittal, Hugh P. Osborn, Beatriz Campos Estrada, James E. Owen, Jessica Libby-Roberts, Arvind F. Gupta, Brad Foley, Erik Meier Vald\'es, Daniel J. Stevens, Ashley Herbst | A Disintegrating Rocky World Shrouded in Dust and Gas: Mid-IR Observations of K2-22b using JWST | Accepted to ApJ Letters on June 3rd, 2025 | null | 10.3847/2041-8213/addfd0 | null | astro-ph.EP | http://creativecommons.org/licenses/by/4.0/ | The disintegrating ultra-short period rocky exoplanet K2-22b periodically emits dusty clouds in a dynamically chaotic process resulting in a variable transit depth from 0-1.3%. The effluents that sublimate off the surface and condense out in space are probably representative of the formerly interior layers convectively transported to the molten surface. Transmission spectroscopy of these transiting clouds reveal spectral fingerprints of the interior composition of this rocky world. We used JWST's Mid-Infrared Instrument (MIRI) as a low-resolution slitless spectrograph to observe four predicted transit windows for K2-22b. For each observation, we extracted a transmission spectrum over the spectral range of 4.4-11.8 $\mu$m. Over the spectral range of 4.4-8 $\mu$m, where the spectral precision is highest, we detect one transit at high significance and two at low significance. While the S/N of the spectrum limits our ability to draw firm conclusions, we find that the data: 1) disfavor featureless, iron-dominated core material, 2) are consistent with some form of magnesium silicate minerals, likely from mantle material, and 3) show a distinct and unexpected feature at $\sim$5 $\mu$m. The unexpected feature, also seen weakly in the low-significance transits, is consistent with an unknown gaseous absorber, possibly NO and/or CO$_2$. These findings warrant further study to improve the constraints on the composition of this disintegrating rocky world. | [
{
"version": "v1",
"created": "Tue, 14 Jan 2025 18:31:15 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 20:52:49 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Tusay",
"Nick",
""
],
[
"Wright",
"Jason T.",
""
],
[
"Beatty",
"Thomas G.",
""
],
[
"Desch",
"Steve",
""
],
[
"Colón",
"Knicole",
""
],
[
"Mittal",
"Tushar",
""
],
[
"Osborn",
"Hugh P.",
""
],
[
"Estrada",
"Beatriz Campos",
""
],
[
"Owen",
"James E.",
""
],
[
"Libby-Roberts",
"Jessica",
""
],
[
"Gupta",
"Arvind F.",
""
],
[
"Foley",
"Brad",
""
],
[
"Valdés",
"Erik Meier",
""
],
[
"Stevens",
"Daniel J.",
""
],
[
"Herbst",
"Ashley",
""
]
] |
2501.08530 | Zhehao Chen | Zhehao Chen, Aslak J J Fellman, Katarzyna Mulewska, Kenichiro Mizohata, Davide Gambino, Yanling Ge, Eryang Lu, Flyura Djurabekova, Andreas Delimitis, Lukasz Kurpaska, Kostas Sarakinos, and Filip Tuomisto | Nanoscale structure formation in nickel-aluminum alloys synthesized far from equilibrium | null | null | null | null | cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The present study reports on the structure formation in thin epitaxial nickel-aluminum films (Ni1-xAlx; Al atomic fraction x up to x=0.24) grown on MgO(001) substrates by magnetron sputtering. Experimental and computational data demonstrate that for x<0.11, the films exhibit the face-centered cubic random solid-solution Ni1-xAlx structure ({\gamma}). Whereas in the range x=0.11-0.24 the phase coexists with the ordered L12 structure ({\gamma}' phase). The two phases are homogenously intermixed forming a coherent and strained nano-solution, which exhibits a single lattice parameter that expands as the Al content increases. Isothermal annealing of films containing x=0.14 of Al, coupled with structural and nano-mechanical characterization, reveal that the nano-solution retains its overall integrity for temperatures up to 673 K, while the film hardness increases from 5.5 GPa (as deposited films) to 6 GPa. Further increase of the annealing temperature to 873 K and 1073 K causes the nano-solution to dissolve into distinct {\gamma} and {\gamma}' phase domains and the hardness to decrease down to values of 4 GPa. These findings confirm the metastable nature of the as-deposited thin Ni1-xAlx alloy films and underpin the effectiveness of high supersaturation/undercooling for creating non-equilibrium phases and self-organized nanostructures upon synthesis of multicomponent materials. | [
{
"version": "v1",
"created": "Wed, 15 Jan 2025 02:39:31 GMT"
},
{
"version": "v2",
"created": "Wed, 22 Jan 2025 13:16:23 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 10:15:12 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chen",
"Zhehao",
""
],
[
"Fellman",
"Aslak J J",
""
],
[
"Mulewska",
"Katarzyna",
""
],
[
"Mizohata",
"Kenichiro",
""
],
[
"Gambino",
"Davide",
""
],
[
"Ge",
"Yanling",
""
],
[
"Lu",
"Eryang",
""
],
[
"Djurabekova",
"Flyura",
""
],
[
"Delimitis",
"Andreas",
""
],
[
"Kurpaska",
"Lukasz",
""
],
[
"Sarakinos",
"Kostas",
""
],
[
"Tuomisto",
"Filip",
""
]
] |
2501.08783 | Shan Cheng | Jian Chai and Shan Cheng | Form factors of light pseudoscalar mesons from the perturbative QCD approach | 46 pages, 14 figures, 6 tables, figure 7 updated, version to appear in JHEP | null | null | null | hep-ph | http://creativecommons.org/licenses/by/4.0/ | We study the electromagnetic and meson-photon transition form factors (TFF) of light pseudoscalar mesons from the perturbative QCD (pQCD) approach. To comprehensively account for both the longitudinal and transverse nonperturbative dynamics of hadronic constituents, we incorpoarate intrinsic transverse momentum distributions (iTMDs) alongside the conventional light-cone distribution amplitudes (LCDAs). The main motivations of this work are the disjointedness of electromagnetic form factors between the theoretical predictions and the experimental measurements, and the BaBar-Belle tension of pion-photon transition form factor in the large momentum transfers. Our calculation is carried out at the next-to-leading-order for the contributions from leading and subleading twist LCDAs, and leading order for the twist four contributions. Notably, this work presents the first systematic evaluation of higher-twist contributions to meson-photon TFFs. The key findings are: (a) iTMDs play a crucial role in describing form factor data, particularly in the small-to-intermediate momentum transfer region where they induce significant modifications to pQCD predictions. (b) The extracted transverse size parameters for valence quark states are found to be $\beta_\pi^2 = 0.51 \pm 0.04$ GeV$^{-2}$ and $\beta_K^2 = 0.30 \pm 0.05$ GeV$^{-2}$, the chiral mass of pion meson $m_0^\pi$ at $1$ GeV is determined to be $1.84 \pm 0.07$ GeV. (c) The meson-photon TFFs are predominantly governed by leading-twist LCDAs. The iTMDs-enhanced pQCD results show better agreement with Belle's pion TFF data across intermediate and large momentum transfers and favor a small $\eta-\eta^\prime$ mixing angle. (d) Remarkably, the inclusion of iTMDs extends the applicability of pQCD calculations down to a few GeV$^2$ for all considered form factors, significantly improving the theory-data consistency. | [
{
"version": "v1",
"created": "Wed, 15 Jan 2025 13:06:53 GMT"
},
{
"version": "v2",
"created": "Tue, 06 May 2025 08:46:41 GMT"
},
{
"version": "v3",
"created": "Wed, 07 May 2025 05:37:43 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 03:16:21 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chai",
"Jian",
""
],
[
"Cheng",
"Shan",
""
]
] |
2501.09023 | Mohammed Shafae | Md Habibor Rahman (1), Mohammed Shafae (2) ((1) University of Massachusetts Dartmouth, (2) The University of Arizona) | Cyber-Physical Security Vulnerabilities Identification and Classification in Smart Manufacturing -- A Defense-in-Depth Driven Framework and Taxonomy | 39 pages (including references), 12 figures | null | 10.1115/1.4068844 | null | cs.CR | http://creativecommons.org/licenses/by-nc-nd/4.0/ | The increasing cybersecurity threats to critical manufacturing infrastructure necessitate proactive strategies for vulnerability identification, classification, and assessment. Traditional approaches, which define vulnerabilities as weaknesses in computational logic or information systems, often overlook the physical and cyber-physical dimensions critical to manufacturing systems, comprising intertwined cyber, physical, and human elements. As a result, existing solutions fall short in addressing the complex, domain-specific vulnerabilities of manufacturing environments. To bridge this gap, this work redefines vulnerabilities in the manufacturing context by introducing a novel characterization based on the duality between vulnerabilities and defenses. Vulnerabilities are conceptualized as exploitable gaps within various defense layers, enabling a structured investigation of manufacturing systems. This paper presents a manufacturing-specific cyber-physical defense-in-depth model, highlighting how security-aware personnel, post-production inspection systems, and process monitoring approaches can complement traditional cyber defenses to enhance system resilience. Leveraging this model, we systematically identify and classify vulnerabilities across the manufacturing cyberspace, human element, post-production inspection systems, production process monitoring, and organizational policies and procedures. This comprehensive classification introduces the first taxonomy of cyber-physical vulnerabilities in smart manufacturing systems, providing practitioners with a structured framework for addressing vulnerabilities at both the system and process levels. Finally, the effectiveness of the proposed model and framework is demonstrated through an illustrative smart manufacturing system and its corresponding threat model. | [
{
"version": "v1",
"created": "Sun, 29 Dec 2024 11:41:06 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 18:40:30 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Rahman",
"Md Habibor",
"",
"University of Massachusetts Dartmouth"
],
[
"Shafae",
"Mohammed",
"",
"The University of Arizona"
]
] |
2501.09500 | Ilja Klebanov | Vesa Kaarnioja, Ilja Klebanov, Claudia Schillings, Yuya Suzuki | Lattice Rules Meet Kernel Cubature | 17 pages, 2 figures | null | null | null | math.NA cs.NA math.ST stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Rank-1 lattice rules are a class of equally weighted quasi-Monte Carlo methods that achieve essentially linear convergence rates for functions in a reproducing kernel Hilbert space (RKHS) characterized by square-integrable first-order mixed partial derivatives. In this work, we explore the impact of replacing the equal weights in lattice rules with optimized cubature weights derived using the reproducing kernel. We establish a theoretical result demonstrating a doubled convergence rate in the one-dimensional case and provide numerical investigations of convergence rates in higher dimensions. We also present numerical results for an uncertainty quantification problem involving an elliptic partial differential equation with a random coefficient. | [
{
"version": "v1",
"created": "Thu, 16 Jan 2025 12:20:42 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 10:49:20 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kaarnioja",
"Vesa",
""
],
[
"Klebanov",
"Ilja",
""
],
[
"Schillings",
"Claudia",
""
],
[
"Suzuki",
"Yuya",
""
]
] |
2501.09622 | Bruno Costa Alves Freire | Bruno C. A. Freire, Nicolas Delfosse, Anthony Leverrier | Optimizing hypergraph product codes with random walks, simulated annealing and reinforcement learning | 6 pages, 2 figures. Parity-check matrices available as text files. Accepted for publication at the IEEE 2025 ISIT | null | null | null | quant-ph cs.IT math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Hypergraph products are quantum low-density parity-check (LDPC) codes constructed from two classical LDPC codes. Although their dimension and distance depend only on the parameters of the underlying classical codes, optimizing their performance against various noise channels remains challenging. This difficulty partly stems from the complexity of decoding in the quantum setting. The standard, ad hoc approach typically involves selecting classical LDPC codes with large girth. In this work, we focus on optimizing performance against the quantum erasure channel. A key advantage of this channel is the existence of an efficient maximum-likelihood decoder, which enables us to employ optimization techniques based on sampling random codes, such as Reinforcement Learning (RL) and Simulated Annealing (SA). Our results indicate that these techniques improve performance relative to the state-of-the-art. | [
{
"version": "v1",
"created": "Thu, 16 Jan 2025 16:01:02 GMT"
},
{
"version": "v2",
"created": "Fri, 17 Jan 2025 06:47:45 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 14:30:00 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Freire",
"Bruno C. A.",
""
],
[
"Delfosse",
"Nicolas",
""
],
[
"Leverrier",
"Anthony",
""
]
] |
2501.09659 | Wei Bu | Wei Bu, Uri Kol, Ziming Liu | Fokker-Planck to Callan-Symanzik: evolution of weight matrices under training | 8 pages, 9 figures | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The dynamical evolution of a neural network during training has been an incredibly fascinating subject of study. First principal derivation of generic evolution of variables in statistical physics systems has proved useful when used to describe training dynamics conceptually, which in practice means numerically solving equations such as Fokker-Planck equation. Simulating entire networks inevitably runs into the curse of dimensionality. In this paper, we utilize Fokker-Planck to simulate the probability density evolution of individual weight matrices in the bottleneck layers of a simple 2-bottleneck-layered auto-encoder and compare the theoretical evolutions against the empirical ones by examining the output data distributions. We also derive physically relevant partial differential equations such as Callan-Symanzik and Kardar-Parisi-Zhang equations from the dynamical equation we have. | [
{
"version": "v1",
"created": "Thu, 16 Jan 2025 16:54:40 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 10:28:30 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Bu",
"Wei",
""
],
[
"Kol",
"Uri",
""
],
[
"Liu",
"Ziming",
""
]
] |
2501.10306 | Sara Veneruso | Michael Herty, Sara Veneruso | Micro-Macro Decomposition of Particle Swarm Optimization Methods | null | null | null | null | math.OC | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Solving non-convex minimization problems using multi-particle metaheuristic derivative-free optimization methods is still an active area of research. Popular methods are Particle Swarm Optimization (PSO) methods, that iteratively update a population of particles according to dynamics inspired by social interactions between individuals. We present a modification to include constrained minimization problems using exact penalization. Additionally, we utilize the hierarchical structure of PSO to introduce a micro-macro decomposition of the algorithm. The probability density of particles is written as a convex combination of microscopic and macroscopic contributions, and both parts are propagated separately. The decomposition is dynamically updated based on heuristic considerations. Numerical examples compare the results obtained using the algorithm in the microscopic scale, in the macroscopic scale, and, using the new micro-macro decomposition. | [
{
"version": "v1",
"created": "Fri, 17 Jan 2025 16:59:08 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 13:50:08 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Herty",
"Michael",
""
],
[
"Veneruso",
"Sara",
""
]
] |
2501.11694 | Aurindam Mondal Mr | Aurindam Mondal, Subir Ghosh | Gravitational Wave Propagation in a Geometric Condensate in Starobinsky Cosmology | Accepted for publication by EPJC. Minor changes have been made. Plots are modified slightly. Title of the manuscript has been changed in an appropriate manner | null | null | null | gr-qc hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper we propose a new paradigm for cosmology: a time dependent scalar condensate background originated from the quadratic $(R + \alpha R^2)$ Starobinski model, where $R$ is the Ricci scalar and $\alpha$ the coupling constant. In weak gravity limit the system decouples into a conventional graviton and a higher derivative scalar. It was shown earlier through works from our group, \cite{ssg,sg,us}, that the latter can sustain an oscillatory lowest energy configuration or a {\it{Geometric Condensate}} as it consists entirely of metric degrees of freedom. In the present work, we study Gravitational Wave propagation in this condensate background. We show that the explicit time dependent nature of the condensate can generate curvature and radiation-like contributions in the scale factor evolution in FLRW cosmology. Subsequently the condensate leaves its signature on the Gravitational Wave profile as it propagates in the condensate modified FLRW spacetime. The wave profile is calculated analytically in terms of Whittaker functions. The main novelty of the Geometric Condensate scheme is that no external (condensate) matter from outside has been considered. | [
{
"version": "v1",
"created": "Mon, 20 Jan 2025 19:17:33 GMT"
},
{
"version": "v2",
"created": "Mon, 03 Feb 2025 09:57:32 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 19:19:09 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Mondal",
"Aurindam",
""
],
[
"Ghosh",
"Subir",
""
]
] |
2501.12227 | Viswanathan Ramachandran | Viswanathan Ramachandran, Tobias J. Oechtering and Mikael Skoglund | Multi-terminal Strong Coordination over Noisy Channels with Encoder Co-operation | Updated Theorem 2. 7 pages, 1 figure. arXiv admin note: substantial text overlap with arXiv:2411.14123 | null | null | null | cs.IT math.IT | http://creativecommons.org/licenses/by/4.0/ | We investigate the problem of strong coordination over a multiple-access channel (MAC) with cribbing encoders. In this configuration, two encoders observe independent and identically distributed (i.i.d.) samples of a source random variable each and encode the inputs to the MAC. The decoder which observes the output of the MAC together with side-information, must generate approximately i.i.d. samples of another random variable which is jointly distributed with the two sources and the side information. We also allow for possible encoder cooperation, where one of the encoders can non-causally crib from the other encoders input. Independent pairwise shared randomness is assumed between each encoder and the decoder at limited rates. Firstly, in the presence of cribbing, we derive an achievable region based on joint source-channel coding. We also prove that in the absence of cribbing, our inner bound is tight for the special case when the MAC is composed of deterministic links, and the sources are conditionally independent given the side information. We then explicitly compute the regions for an example both with and without cribbing between the encoders, and demonstrate that cribbing strictly improves upon the achievable region. | [
{
"version": "v1",
"created": "Tue, 21 Jan 2025 15:51:36 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 14:16:43 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ramachandran",
"Viswanathan",
""
],
[
"Oechtering",
"Tobias J.",
""
],
[
"Skoglund",
"Mikael",
""
]
] |
2501.12729 | Vo Phuc Dang | Dang Vo Phuc | On the algebraic transfers of ranks 4 and 6 at generic degrees | 31 pages. This version corrects a typo in the Introduction and several computational errors in the proof of Theorem 1.2 (Part I), as well as notation inconsistencies involving the parameters $s$ and $r$ in Part II. These updates do not affect the validity of the main theorems. A full conflict of interest disclosure is included | Rendiconti del Circolo Matematico di Palermo Series 2, Vol.74 (Article number 38):1-33, 2025 | 10.1007/s12215-024-01141-0 | null | math.AT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Let $\mathscr A$ denote the classical singly-graded Steenrod algebra over the binary field $\mathbb Z/2.$ We write $P_k:=\mathbb Z/2[t_1, t_2, \ldots, t_k]$ as the polynomial algebra on $k$ generators, each having a degree of one. Let $GL_k$ be the general linear group of rank $k$ over $\mathbb Z/2.$ Then, $P_k$ is an $ \mathscr A[GL_k]$-module. The structure of the cohomology groups, ${\rm Ext}_{ \mathscr A}^{k, k+\bullet}(\mathbb Z/2, \mathbb Z/2)$, of the Steenrod algebra has, thus far, resisted clear understanding and full description for all homological degrees $k$. In the study of these groups, the algebraic transfer -- constructed by W. Singer in [Math. Z. 202, 493--523 (1989)] -- plays an important role. The Singer transfer is represented by the following homomorphism: $$Tr_k: {\rm Hom}([(\mathbb Z/2\otimes_{ \mathscr A} P_k)_{\bullet}]^{GL_k}, \mathbb Z/2)\longrightarrow {\rm Ext}_{ \mathscr A}^{k, k+\bullet}(\mathbb Z/2, \mathbb Z/2).$$ Among Singer's contributions is an interesting open conjecture asserting the monomorphism of $Tr_k$ for all $k.$ For this reason, our main aim in this article is to ascertain the validity of the Singer conjecture for ranks 4 and 6 in certain families of internal degrees. We place particular emphasis on the rank 4 case. More precisely, we present a detailed proof for certain generic degree cases when verifying the conjecture of rank four, which were succinctly noted in our previous work [Proc. Roy. Soc. Edinburgh Sect. A 153, 1529--1542 (2023)]. | [
{
"version": "v1",
"created": "Wed, 22 Jan 2025 09:05:13 GMT"
},
{
"version": "v2",
"created": "Fri, 24 Jan 2025 14:28:07 GMT"
},
{
"version": "v3",
"created": "Tue, 28 Jan 2025 13:17:56 GMT"
},
{
"version": "v4",
"created": "Tue, 03 Jun 2025 14:49:35 GMT"
},
{
"version": "v5",
"created": "Thu, 05 Jun 2025 11:41:37 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Phuc",
"Dang Vo",
""
]
] |
2501.12823 | Hilmy Baja | Hilmy Baja, Michiel Kallenberg, Ioannis N. Athanasiadis | To Measure or Not: A Cost-Sensitive, Selective Measuring Environment for Agricultural Management Decisions with Reinforcement Learning | 10 pages, 3 figures, accepted after peer-review at the 39th Annual AAAI Conference on Artificial Intelligence, AI for Social Impact Track, February 2025, Philadelphia, Pennsylvania, USA | null | 10.1609/aaai.v39i27.34999 | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Farmers rely on in-field observations to make well-informed crop management decisions to maximize profit and minimize adverse environmental impact. However, obtaining real-world crop state measurements is labor-intensive, time-consuming and expensive. In most cases, it is not feasible to gather crop state measurements before every decision moment. Moreover, in previous research pertaining to farm management optimization, these observations are often assumed to be readily available without any cost, which is unrealistic. Hence, enabling optimization without the need to have temporally complete crop state observations is important. An approach to that problem is to include measuring as part of decision making. As a solution, we apply reinforcement learning (RL) to recommend opportune moments to simultaneously measure crop features and apply nitrogen fertilizer. With realistic considerations, we design an RL environment with explicit crop feature measuring costs. While balancing costs, we find that an RL agent, trained with recurrent PPO, discovers adaptive measuring policies that follow critical crop development stages, with results aligned by what domain experts would consider a sensible approach. Our results highlight the importance of measuring when crop feature measurements are not readily available. | [
{
"version": "v1",
"created": "Wed, 22 Jan 2025 12:03:53 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Baja",
"Hilmy",
""
],
[
"Kallenberg",
"Michiel",
""
],
[
"Athanasiadis",
"Ioannis N.",
""
]
] |
2501.12911 | Abdulkadir Korkmaz | Abdulkadir Korkmaz and Praveen Rao | A Selective Homomorphic Encryption Approach for Faster Privacy-Preserving Federated Learning | 18 pages, 18 figures | null | null | null | cs.CR cs.DC cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Federated learning (FL) has come forward as a critical approach for privacy-preserving machine learning in healthcare, allowing collaborative model training across decentralized medical datasets without exchanging clients' data. However, current security implementations for these systems face a fundamental trade-off: rigorous cryptographic protections like fully homomorphic encryption (FHE) impose prohibitive computational overhead, while lightweight alternatives risk vulnerable data leakage through model updates. To address this issue, we present FAS (Fast and Secure Federated Learning), a novel approach that strategically combines selective homomorphic encryption, differential privacy, and bitwise scrambling to achieve robust security without compromising practical usability. Our approach eliminates the need for model pretraining phases while dynamically protecting high-risk model parameters through layered encryption and obfuscation. We implemented FAS using the Flower framework and evaluated it on a cluster of eleven physical machines. Our approach was up to 90\% faster than applying FHE on the model weights. In addition, we eliminated the computational overhead that is required by competitors such as FedML-HE and MaskCrypt. Our approach was up to 1.5$\times$ faster than the competitors while achieving comparable security results.
Experimental evaluations on medical imaging datasets confirm that FAS maintains similar security results to conventional FHE against gradient inversion attacks while preserving diagnostic model accuracy. These results position FAS as a practical solution for latency-sensitive healthcare applications where both privacy preservation and computational efficiency are requirements. | [
{
"version": "v1",
"created": "Wed, 22 Jan 2025 14:37:44 GMT"
},
{
"version": "v2",
"created": "Thu, 27 Feb 2025 21:23:54 GMT"
},
{
"version": "v3",
"created": "Thu, 27 Mar 2025 17:44:27 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 13:47:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Korkmaz",
"Abdulkadir",
""
],
[
"Rao",
"Praveen",
""
]
] |
2501.13235 | Daniela Ballari | Daniela Ballari, Willington Siabato, Christophe Claramunt, Felix Mata, Roberto Zagal, Rodolfo Franco | On the development of open geographical data infrastructures in Latin America: progress and challenges | 10 pages, 2 figures | 2025 | 10.5311/JOSIS.2025.30.364 | 30 | cs.CY | http://creativecommons.org/licenses/by/4.0/ | Open data initiatives and infrastructures play an essential role in favoring better data access, participation, and transparency in government operations and decision-making. Open Geographical Data Infrastructures (OGDIs) allow citizens to access and scrutinize government and public data, thereby enhancing accountability and evidence-based decision-making. This encourages citizen engagement and participation in public affairs and offers researchers, non-governmental organizations, civil society, and business sectors novel opportunities to analyze and disseminate large amounts of geographical data and to address social, urban, and environmental challenges. In Latin America, while recent open government agendas have shown an inclination towards transparency, citizen participation, and collaboration, only a limited number of OGDIs allow unrestricted use and re-use of their data. Given the region's cultural, social, and economic disparities, there is a contrasting digital divide that significantly impacts how OGDIs are being developed. Therefore, this paper analyses recent progress in developing OGDIs in Latin America, technological gaps, and open geographical data initiatives. The main results denote an early development of OGDIs in the region. Nevertheless, this opens the door for the timely involvement of citizens and non-government sectors to share needs, experiences, knowledge, and expertise, as well as to address a transboundary research agenda. Challenges are discussed from multiple perspectives: data, methodological, governmental and readiness, and potential impact. This analysis is aimed at researchers, policymakers, and practitioners interested in the specific challenges and progress of OGDIs in Latin America, while also contributing to the global conversation on best practices and lessons learned in implementing OGDIs across different contexts. | [
{
"version": "v1",
"created": "Wed, 22 Jan 2025 21:37:40 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ballari",
"Daniela",
""
],
[
"Siabato",
"Willington",
""
],
[
"Claramunt",
"Christophe",
""
],
[
"Mata",
"Felix",
""
],
[
"Zagal",
"Roberto",
""
],
[
"Franco",
"Rodolfo",
""
]
] |
2501.14244 | Satori Tsuzuki Ph.D | Satori Tsuzuki | Multi-scale physics of cryogenic liquid helium-4: Inverse coarse-graining properties of smoothed particle hydrodynamics | null | null | null | null | physics.flu-dyn cond-mat.mes-hall cond-mat.other physics.comp-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Our recent numerical studies on cryogenic liquid helium-4 highlight key features of multiscale physics that can be captured using the two-fluid model. In this paper, we demonstrated that classical and quantum hydrodynamic two-fluid models are connected via scale transformations: large eddy simulation (LES) filtering links microscopic to macroscopic scales, while inverse scale transformation through SPH connects macro back to microscales. We showed that the spin angular momentum conservation term, introduced as a quantum-like correction, formally corresponds to a subgrid-scale (SGS) model derived from this transformation. Moreover, solving the classical hydrodynamic two-fluid model with SPH appears to reproduce microscopic-scale fluctuations at macroscopic scales. The amplitude of these fluctuations depends on the kernel radius. This effect may arise from truncation errors from kernel smoothing, which can qualitatively resemble such fluctuations. However, this resemblance lacks first-principle justification and should be viewed as a speculative analogy rather than a physically grounded effect. Our theoretical analysis further suggests that the Condiff viscosity model can act as an SGS model, incorporating quantum vortex interactions under point-vortex approximation into the two-fluid framework. These findings provide new insight into the microscopic structure of cryogenic helium-4 within a multiscale context. Notably, the normal fluid can be understood as a mixture of inviscid and viscous fluid particles. While molecular viscosity renders the normal fluid at microscopic scales, its small magnitude contributes little to the large-scale effective viscosity, which includes both molecular and eddy viscosities; therefore, in laminar regimes where eddy viscosity is negligible, the normal fluid may be effectively treated as inviscid at large scales if molecular viscosity is sufficiently small. | [
{
"version": "v1",
"created": "Fri, 24 Jan 2025 05:09:58 GMT"
},
{
"version": "v2",
"created": "Mon, 24 Feb 2025 18:31:21 GMT"
},
{
"version": "v3",
"created": "Tue, 25 Feb 2025 20:00:57 GMT"
},
{
"version": "v4",
"created": "Sat, 01 Mar 2025 18:20:43 GMT"
},
{
"version": "v5",
"created": "Mon, 10 Mar 2025 17:43:14 GMT"
},
{
"version": "v6",
"created": "Fri, 23 May 2025 19:25:06 GMT"
},
{
"version": "v7",
"created": "Thu, 05 Jun 2025 11:42:16 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Tsuzuki",
"Satori",
""
]
] |
2501.14429 | Krist\'of Kanalas | Krist\'of Kanalas | Various topos of types constructions | null | null | null | null | math.CT math.LO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study and compare some topos of types constructions, which were defined by Garner, Joyal, Reyes and Makkai. | [
{
"version": "v1",
"created": "Fri, 24 Jan 2025 11:56:30 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 11:30:48 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kanalas",
"Kristóf",
""
]
] |
2501.16642 | Yixuan Jia | Siyi Chen, Yixuan Jia, Qing Qu, He Sun, Jeffrey A Fessler | FlowDAS: A Stochastic Interpolant-based Framework for Data Assimilation | null | null | null | null | eess.SP cs.LG eess.IV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Data assimilation (DA) integrates observations with a dynamical model to estimate states of PDE-governed systems. Model-driven methods (e.g., Kalman, particle) presuppose full knowledge of the true dynamics, which is not always satisfied in practice, while purely data-driven solvers learn a deterministic mapping between observations and states and therefore miss the intrinsic stochasticity of real processes. Recently, score-based diffusion models learn a global diffusion prior and provide a good modeling of the stochastic dynamics, showing new potential for DA. However, their all-at-once generation rather than step-by-step transition limits their performance when dealing with highly complex stochastic processes and lacks physical interpretability. To tackle these drawbacks, we introduce FlowDAS, a generative DA framework that uses stochastic interpolants to directly learn state transition dynamics and achieve step-by-step transition to better model the real dynamics. We also improve the framework by combining the observation, better suiting the DA settings. Directly learning the underlying dynamics from collected data removes restrictive dynamical assumptions, and conditioning on observations at each interpolation step yields stable, measurement-consistent forecasts. Experiments on Lorenz-63, Navier-Stokes super-resolution/sparse-observation scenarios, and large-scale weather forecasting -- where dynamics are partly or wholly unknown -- show that FlowDAS surpasses model-driven methods, neural operators, and score-based baselines in accuracy and physical plausibility. | [
{
"version": "v1",
"created": "Mon, 13 Jan 2025 05:03:41 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 07:58:03 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chen",
"Siyi",
""
],
[
"Jia",
"Yixuan",
""
],
[
"Qu",
"Qing",
""
],
[
"Sun",
"He",
""
],
[
"Fessler",
"Jeffrey A",
""
]
] |
2501.17753 | Ruben Gargiulo | Ruben Gargiulo, Elisa Di Meco, Stefano Palmisano | Feasibility study of true muonium observation with the existing Belle-II dataset | Version accepted for publication in PhysRevD | null | null | null | hep-ph hep-ex | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | True muonium ($\mu^+\mu^-$) is one of the cleanest bound states, being composed only of leptons, along with true tauonium and positronium. Unlike the latter, true muonium and true tauonium have not been observed so far. This article shows that the spin-0 state of true muonium (para-TM), decaying into two photons, can be observed at a discovery level of significance in the dataset already collected by the Belle-II experiment at the $\Upsilon (4S)$ peak, with certain assumptions on systematic uncertainties. Para-TM is produced via photon-photon fusion, and its observation is based on the detection of the photon pair resulting from its decay, on top of the continuum background due predominantly to light-by-light scattering. Trigger, acceptance and isolation cuts, along with calorimeter resolution and reconstruction efficiency, are taken into account during the Monte Carlo simulation of both signal and background. In order to separate signal and background, a machine learning method, based on extremely randomized trees, is trained on simulated events. Finally, the expected statistical significance of TM observation is evaluated, taking into account systematic uncertainties in a parametric fashion. | [
{
"version": "v1",
"created": "Wed, 29 Jan 2025 16:47:56 GMT"
},
{
"version": "v2",
"created": "Fri, 04 Apr 2025 12:20:43 GMT"
},
{
"version": "v3",
"created": "Mon, 07 Apr 2025 07:57:56 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 07:01:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Gargiulo",
"Ruben",
""
],
[
"Di Meco",
"Elisa",
""
],
[
"Palmisano",
"Stefano",
""
]
] |
2501.18527 | Christoph Spiegel | Konrad Mundinger, Max Zimmer, Aldo Kiem, Christoph Spiegel, Sebastian Pokutta | Neural Discovery in Mathematics: Do Machines Dream of Colored Planes? | 9 pages main paper, 11 pages references and appendix, 17 figures, 1 table | Proc. 42nd ICML, PMLR 267, 2025 | null | null | cs.LG math.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We demonstrate how neural networks can drive mathematical discovery through a case study of the Hadwiger-Nelson problem, a long-standing open problem at the intersection of discrete geometry and extremal combinatorics that is concerned with coloring the plane while avoiding monochromatic unit-distance pairs. Using neural networks as approximators, we reformulate this mixed discrete-continuous geometric coloring problem with hard constraints as an optimization task with a probabilistic, differentiable loss function. This enables gradient-based exploration of admissible configurations that most significantly led to the discovery of two novel six-colorings, providing the first improvement in thirty years to the off-diagonal variant of the original problem. Here, we establish the underlying machine learning approach used to obtain these results and demonstrate its broader applicability through additional numerical insights. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 17:44:34 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 09:30:22 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 07:01:14 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Mundinger",
"Konrad",
""
],
[
"Zimmer",
"Max",
""
],
[
"Kiem",
"Aldo",
""
],
[
"Spiegel",
"Christoph",
""
],
[
"Pokutta",
"Sebastian",
""
]
] |
2501.18694 | Lucas Maisel Licer\'an | L. Maisel Licer\'an and H. T. C. Stoof | Unconventional excitonic insulators in two-dimensional topological materials | null | Maisel Licer\'an, L., & Stoof, H. T. C. (2025). Physical Review B, 111(24), 245102 | 10.1103/PhysRevB.111.245102 | null | cond-mat.str-el cond-mat.mes-hall cond-mat.mtrl-sci cond-mat.supr-con quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Bound electron-hole pairs in semiconductors known as excitons can form a coherent state at low temperatures akin to a BCS condensate. The resulting phase is known as the excitonic insulator and has superfluid properties. Here we theoretically study the excitonic insulator in a pair of recently proposed two-dimensional candidate materials with nontrivial band topology. Contrary to previous works, we include interaction channels that violate the individual electron and hole number conservations. These are on equal footing with the number-conserving processes due to the substantial overlap of Wannier orbitals of different bands, which cannot be exponentially localized due to the nontrivial Chern numbers of the latter. Their inclusion is crucial to determine the symmetry of the electron-hole pairing, and by performing mean-field calculations at nonzero temperatures we find that the order parameter in these systems is a chiral $d$-wave. We discuss the nontrivial topology of this unconventional state and discuss some properties of the associated Berezinskii-Kosterlitz-Thouless transition. In particular, we argue that here it becomes a smooth crossover and estimate the associated temperature to lie between 50 K and 75 K on realistic substrates, over an order of magnitude larger than in the number-conserving approximation where $s$-wave pairing is favored. Our results highlight the interplay between topology at the single-particle level and long-range interactions, motivating further research in systems where both phenomena coexist. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 19:00:06 GMT"
},
{
"version": "v2",
"created": "Thu, 08 May 2025 14:54:36 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 08:25:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Licerán",
"L. Maisel",
""
],
[
"Stoof",
"H. T. C.",
""
]
] |
2501.18878 | Yixuan Guo | Yixuan Guo, Shuaifan Xia, Mingliang Xiong, Qingwen Liu, Wen Fang, Qingwei Jiang, Gang Yan, Jiangchuan Mu | Integrated Sensing and Communication System Based on Radio Frequency Resonance Beam | null | null | null | null | eess.SP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | To address the complex beam control in traditional multiple-input multiple-output (MIMO) systems, researchers have proposed adaptive beam alignment using retro-directive antenna (RDA) arrays. This approach creates echo resonance between the base station (BS) and user equipment (UE), significantly reducing computational load. However, conventional resonant beam systems (RBS) suffer from echo interference due to the shared uplink and downlink frequency. Therefore, this paper proposes an innovative resonance beam-based integrated sensing and communication (RB-ISAC) system designed for efficient passive sensing and bidirectional communication. In this system, the UE operates passively, with both the BS and UE utilizing a phase conjugation and frequency conversion structure to decouple uplink and downlink carrier frequencies, ensuring continuous electromagnetic wave oscillation between the two ends. Effective compensation for signal propagation loss enables resonance after multiple oscillations. At this point, the beam's field forms a low-diffraction-loss, highly focused pattern, automatically aligning the transmitter and receiver. This enables high-precision passive positioning alongside robust uplink and downlink communication. Simulation results demonstrate the proposed system achieves resonance within multiple iterations, supporting uplink and downlink communication up to 5 m, and enabling passive direction of arrival (DOA) estimation with an error under 2$^\circ$ . | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 04:16:30 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 05:44:50 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Guo",
"Yixuan",
""
],
[
"Xia",
"Shuaifan",
""
],
[
"Xiong",
"Mingliang",
""
],
[
"Liu",
"Qingwen",
""
],
[
"Fang",
"Wen",
""
],
[
"Jiang",
"Qingwei",
""
],
[
"Yan",
"Gang",
""
],
[
"Mu",
"Jiangchuan",
""
]
] |
2501.19054 | Ruiyu Wang | Ruiyu Wang, Yu Yuan, Shizhao Sun, Jiang Bian | Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models | ICML 2025 camera ready | null | null | null | cs.CV cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Creating Computer-Aided Design (CAD) models requires significant expertise and effort. Text-to-CAD, which converts textual descriptions into CAD parametric sequences, is crucial in streamlining this process. Recent studies have utilized ground-truth parametric sequences, known as sequential signals, as supervision to achieve this goal. However, CAD models are inherently multimodal, comprising parametric sequences and corresponding rendered visual objects. Besides,the rendering process from parametric sequences to visual objects is many-to-one. Therefore, both sequential and visual signals are critical for effective training. In this work, we introduce CADFusion, a framework that uses Large Language Models (LLMs) as the backbone and alternates between two training stages: the sequential learning (SL) stage and the visual feedback (VF) stage. In the SL stage, we train LLMs using ground-truth parametric sequences, enabling the generation of logically coherent parametric sequences. In the VF stage, we reward parametric sequences that render into visually preferred objects and penalize those that do not, allowing LLMs to learn how rendered visual objects are perceived and evaluated. These two stages alternate throughout the training, ensuring balanced learning and preserving benefits of both signals. Experiments demonstrate that CADFusion significantly improves performance, both qualitatively and quantitatively. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 11:28:16 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Feb 2025 10:43:26 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 08:25:25 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wang",
"Ruiyu",
""
],
[
"Yuan",
"Yu",
""
],
[
"Sun",
"Shizhao",
""
],
[
"Bian",
"Jiang",
""
]
] |
2501.19158 | Giovanni Catania | Giovanni Catania, Aur\'elien Decelle, Cyril Furtlehner, Beatriz Seoane | A theoretical framework for overfitting in energy-based modeling | 29 pages, 20 figures (including appendix). Accepted at Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025 | null | null | null | cs.LG cond-mat.dis-nn cond-mat.stat-mech | http://creativecommons.org/licenses/by-nc-nd/4.0/ | We investigate the impact of limited data on training pairwise energy-based models for inverse problems aimed at identifying interaction networks. Utilizing the Gaussian model as testbed, we dissect training trajectories across the eigenbasis of the coupling matrix, exploiting the independent evolution of eigenmodes and revealing that the learning timescales are tied to the spectral decomposition of the empirical covariance matrix. We see that optimal points for early stopping arise from the interplay between these timescales and the initial conditions of training. Moreover, we show that finite data corrections can be accurately modeled through asymptotic random matrix theory calculations and provide the counterpart of generalized cross-validation in the energy based model context. Our analytical framework extends to binary-variable maximum-entropy pairwise models with minimal variations. These findings offer strategies to control overfitting in discrete-variable models through empirical shrinkage corrections, improving the management of overfitting in energy-based generative models. Finally, we propose a generalization to arbitrary energy-based models by deriving the neural tangent kernel dynamics of the score function under the score-matching algorithm. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 14:21:02 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 15:50:19 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 17:06:46 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Catania",
"Giovanni",
""
],
[
"Decelle",
"Aurélien",
""
],
[
"Furtlehner",
"Cyril",
""
],
[
"Seoane",
"Beatriz",
""
]
] |
2501.19182 | Rafael Tov Elberg Haschelevici | Rafael Elberg, Felipe del Rio, Mircea Petrache, Denis Parra | A Compressive-Expressive Communication Framework for Compositional Representations | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Compositional generalization--the ability to interpret novel combinations of familiar elements--is a hallmark of human cognition and language. Despite recent advances, deep neural networks still struggle to acquire this property reliably. In this work, we introduce CELEBI (Compressive-Expressive Language Emergence through a discrete Bottleneck and Iterated learning), a novel self-supervised framework for inducing compositionality in learned representations from pre-trained models, through a reconstruction-based communication game between a sender and a receiver. Building on theories of language emergence, we integrate three mechanisms that jointly promote compressibility, expressivity, and efficiency in the emergent language. First, interactive decoding incentivizes intermediate reasoning by requiring the receiver to produce partial reconstructions after each symbol. Second, a reconstruction-based imitation phase, inspired by iterated learning, trains successive generations of agents to imitate reconstructions rather than messages, enforcing a tighter communication bottleneck. Third, pairwise distance maximization regularizes message diversity by encouraging high distances between messages, with formal links to entropy maximization. Our method significantly improves both the efficiency and compositionality of the learned messages on the Shapes3D and MPI3D datasets, surpassing prior discrete communication frameworks in both reconstruction accuracy and topographic similarity. This work provides new theoretical and empirical evidence for the emergence of structured, generalizable communication protocols from simplicity-based inductive biases. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 14:46:11 GMT"
},
{
"version": "v2",
"created": "Thu, 13 Feb 2025 21:04:44 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 13:57:28 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Elberg",
"Rafael",
""
],
[
"del Rio",
"Felipe",
""
],
[
"Petrache",
"Mircea",
""
],
[
"Parra",
"Denis",
""
]
] |
2502.00278 | Tyler Parsotan | Tyler Parsotan, David M. Palmer, Samuele Ronchini, James Delaunay, Aaron Tohuvavohu, Sibasish Laha, Amy Lien, S. Bradley Cenko, Hans Krimm, Craig Markwardt | BatAnalysis -- A Comprehensive Python Pipeline for Swift BAT Time-Tagged Event Data Analysis | 13 pages, 8 figures, accepted to ApJ, BatAnalysis github link is: https://github.com/parsotat/BatAnalysis | null | null | null | astro-ph.HE astro-ph.IM | http://creativecommons.org/licenses/by/4.0/ | The Swift Burst Alert Telescope (BAT) is a coded aperture gamma-ray instrument with a large field of view that was designed to detect and localize transient events. When a transient is detected, either on-board or externally, the BAT saves time-tagged event (TTE) data which provides the highest quality information of the locations of the photons on the detector plane and their energies. This data can be used to produce spectra, lightcurves, and sky images of a transient event. While these data products are produced by the Swift Data Center and can be produced by current software, they are often preset to certain time and energy intervals which has limited their use in the current time domain and multi-messenger environment. Here, we introduce a new capability for the BatAnalysis python package to download and process TTE data under an open-source pythonic framework that allows for easy interfacing with other python packages. The new capabilities of the BatAnalysis software allows for TTE data to be used by the community in a variety of advanced customized analyses of astrophysical sources which BAT may have TTE data for, such as Fast Radio Bursts (FRBs), Gamma-ray Bursts (GRBs), Low Mass X-ray Binaries (LMXB), Soft Gamma Repeaters, magnetars, and many other sources. We highlight the usefulness of the BatAnalysis package in analyzing TTE data produced by an on-board GRB trigger, a FRB external trigger, a sub-threshold detection of the LMXB EXO 0748-676, and an external trigger of a GRB that BAT detected during a slew. | [
{
"version": "v1",
"created": "Sat, 01 Feb 2025 02:28:31 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:33:50 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Parsotan",
"Tyler",
""
],
[
"Palmer",
"David M.",
""
],
[
"Ronchini",
"Samuele",
""
],
[
"Delaunay",
"James",
""
],
[
"Tohuvavohu",
"Aaron",
""
],
[
"Laha",
"Sibasish",
""
],
[
"Lien",
"Amy",
""
],
[
"Cenko",
"S. Bradley",
""
],
[
"Krimm",
"Hans",
""
],
[
"Markwardt",
"Craig",
""
]
] |
2502.00401 | Karish Grover | Karish Grover, Haiyang Yu, Xiang Song, Qi Zhu, Han Xie, Vassilis N. Ioannidis, Christos Faloutsos | Spectro-Riemannian Graph Neural Networks | ICLR 2025 | null | null | null | cs.LG cs.AI stat.ML | http://creativecommons.org/licenses/by/4.0/ | Can integrating spectral and curvature signals unlock new potential in graph representation learning? Non-Euclidean geometries, particularly Riemannian manifolds such as hyperbolic (negative curvature) and spherical (positive curvature), offer powerful inductive biases for embedding complex graph structures like scale-free, hierarchical, and cyclic patterns. Meanwhile, spectral filtering excels at processing signal variations across graphs, making it effective in homophilic and heterophilic settings. Leveraging both can significantly enhance the learned representations. To this end, we propose Spectro-Riemannian Graph Neural Networks (CUSP) - the first graph representation learning paradigm that unifies both CUrvature (geometric) and SPectral insights. CUSP is a mixed-curvature spectral GNN that learns spectral filters to optimize node embeddings in products of constant-curvature manifolds (hyperbolic, spherical, and Euclidean). Specifically, CUSP introduces three novel components: (a) Cusp Laplacian, an extension of the traditional graph Laplacian based on Ollivier-Ricci curvature, designed to capture the curvature signals better; (b) Cusp Filtering, which employs multiple Riemannian graph filters to obtain cues from various bands in the eigenspectrum; and (c) Cusp Pooling, a hierarchical attention mechanism combined with a curvature-based positional encoding to assess the relative importance of differently curved substructures in our graph. Empirical evaluation across eight homophilic and heterophilic datasets demonstrates the superiority of CUSP in node classification and link prediction tasks, with a gain of up to 5.3% over state-of-the-art models. The code is available at: https://github.com/amazon-science/cusp. | [
{
"version": "v1",
"created": "Sat, 01 Feb 2025 11:31:01 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 19:39:52 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Grover",
"Karish",
""
],
[
"Yu",
"Haiyang",
""
],
[
"Song",
"Xiang",
""
],
[
"Zhu",
"Qi",
""
],
[
"Xie",
"Han",
""
],
[
"Ioannidis",
"Vassilis N.",
""
],
[
"Faloutsos",
"Christos",
""
]
] |
2502.00620 | Yihao Xue | Yihao Xue, Jiping Li, Baharan Mirzasoleiman | Representations Shape Weak-to-Strong Generalization: Theoretical Insights and Empirical Predictions | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Weak-to-Strong Generalization (W2SG), where a weak model supervises a stronger one, serves as an important analogy for understanding how humans might guide superhuman intelligence in the future. Promising empirical results revealed that a strong model can surpass its weak supervisor. While recent work has offered theoretical insights into this phenomenon, a clear understanding of the interactions between weak and strong models that drive W2SG remains elusive. We investigate W2SG through a theoretical lens and show that it can be characterized using kernels derived from the principal components of weak and strong models' internal representations. These kernels can be used to define a space that, at a high level, captures what the weak model is unable to learn but is learnable by the strong model. The projection of labels onto this space quantifies how much the strong model falls short of its full potential due to weak supervision. This characterization also provides insights into how certain errors in weak supervision can be corrected by the strong model, regardless of overfitting. Our theory has significant practical implications, providing a representation-based metric that predicts W2SG performance trends without requiring labels, as shown in experiments on molecular predictions with transformers and 5 NLP tasks involving 52 LLMs. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 01:11:51 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Feb 2025 00:36:00 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 20:08:20 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Xue",
"Yihao",
""
],
[
"Li",
"Jiping",
""
],
[
"Mirzasoleiman",
"Baharan",
""
]
] |
2502.00722 | Youhe Jiang | Youhe Jiang, Fangcheng Fu, Xiaozhe Yao, Guoliang He, Xupeng Miao, Ana Klimovic, Bin Cui, Binhang Yuan, Eiko Yoneki | Demystifying Cost-Efficiency in LLM Serving over Heterogeneous GPUs | null | null | null | null | cs.DC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent advancements in Large Language Models (LLMs) have led to increasingly diverse requests, accompanied with varying resource (compute and memory) demands to serve them. However, this in turn degrades the cost-efficiency of LLM serving as common practices primarily rely on homogeneous GPU resources. In response to this problem, this work conducts a thorough study about serving LLMs over heterogeneous GPU resources on cloud platforms. The rationale is that different GPU types exhibit distinct compute and memory characteristics, aligning well with the divergent resource demands of diverse requests. Particularly, through comprehensive benchmarking, we discover that the cost-efficiency of LLM serving can be substantially optimized by meticulously determining GPU composition, deployment configurations, and workload assignments. Subsequently, we design a scheduling algorithm via mixed-integer linear programming, aiming at deducing the most cost-efficient serving plan under the constraints of price budget and real-time GPU availability. Remarkably, our approach effectively outperforms homogeneous and heterogeneous baselines under a wide array of scenarios, covering diverse workload traces, varying GPU availablilities, and multi-model serving. This casts new light on more accessible and efficient LLM serving over heterogeneous cloud resources. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 08:44:43 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 07:54:05 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jiang",
"Youhe",
""
],
[
"Fu",
"Fangcheng",
""
],
[
"Yao",
"Xiaozhe",
""
],
[
"He",
"Guoliang",
""
],
[
"Miao",
"Xupeng",
""
],
[
"Klimovic",
"Ana",
""
],
[
"Cui",
"Bin",
""
],
[
"Yuan",
"Binhang",
""
],
[
"Yoneki",
"Eiko",
""
]
] |
2502.00805 | Shantanu Desai | Vyaas Ramakrishnan, Shantanu Desai | Constraints on Lorentz Invariance Violation from Gamma-ray Burst rest-frame spectral lags using Profile Likelihood | 9 pages, 6 figures. Accepted in Universe | null | null | null | astro-ph.HE astro-ph.IM | http://creativecommons.org/licenses/by/4.0/ | We reanalyze the spectral lag data for 56 Gamma-Ray Bursts (GRBs) in the cosmological rest frame to search for Lorentz Invariance Violation (LIV) using frequentist inference. For this purpose, we use the technique of profile likelihood to deal with the nuisance parameters, corresponding to a constant time lag in the GRB rest frame and an unknown intrinsic scatter, while the parameter of interest is the energy scale for LIV ($E_{QG}$). With this method, we do not obtain a global minimum for $\chi^2$ as a function of $E_{QG}$ up to the Planck scale. Thus, we can obtain one-sided lower limits on $E_{QG}$ in a seamless manner. Therefore, the 95\% c.l. lower limits which we thus obtain on $E_{QG}$ are then given by: $E_{QG}\geq 2.07 \times 10^{14}$ GeV and $E_{QG}\geq 3.71\times 10^{5}$ GeV, for linear and quadratic LIV, respectively. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 13:59:07 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 03:17:22 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ramakrishnan",
"Vyaas",
""
],
[
"Desai",
"Shantanu",
""
]
] |
2502.00837 | Hadi Mohammadi | Hadi Mohammadi, Ayoub Bagheri, Anastasia Giachanou, and Daniel L. Oberski | Explainability in Practice: A Survey of Explainable NLP Across Various Domains | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Natural Language Processing (NLP) has become a cornerstone in many critical sectors, including healthcare, finance, and customer relationship management. This is especially true with the development and use of advanced models such as GPT-based architectures and BERT, which are widely used in decision-making processes. However, the black-box nature of these advanced NLP models has created an urgent need for transparency and explainability. This review explores explainable NLP (XNLP) with a focus on its practical deployment and real-world applications, examining its implementation and the challenges faced in domain-specific contexts. The paper underscores the importance of explainability in NLP and provides a comprehensive perspective on how XNLP can be designed to meet the unique demands of various sectors, from healthcare's need for clear insights to finance's emphasis on fraud detection and risk assessment. Additionally, this review aims to bridge the knowledge gap in XNLP literature by offering a domain-specific exploration and discussing underrepresented areas such as real-world applicability, metric evaluation, and the role of human interaction in model assessment. The paper concludes by suggesting future research directions that could enhance the understanding and broader application of XNLP. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 16:18:44 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:41:25 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Mohammadi",
"Hadi",
""
],
[
"Bagheri",
"Ayoub",
""
],
[
"Giachanou",
"Anastasia",
""
],
[
"Oberski",
"Daniel L.",
""
]
] |
2502.00875 | Sagar Pokharel | Sagar Pokharel and Albina Tropina | Characteristics of Femtosecond Laser Induced Filament and Energy Coupling by Nanosecond Laser Pulse in Air | null | J. Appl. Phys. 137, 213303 (2025) | 10.1063/5.0263315 | null | physics.plasm-ph physics.app-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This study presents a detailed plasma kinetics model for laser-induced non-equilibrium plasmas in atmospheric pressure air, incorporating a self-consistent energy balance and refined rate expressions within a three-temperature framework. The model is validated against experimental data of femtosecond-laser-induced filaments, showing good agreement in electron dynamics and gas temperature. The analysis focuses on femtosecond filament decay kinetics and characteristic properties across varying initial electron densities and electron temperatures, including cases with oxygen addition and its influence on decay behavior. The study further examines energy coupling between the femtosecond filament and nanosecond laser pulse, identifying dominant kinetic pathways and optimal time delays through a comparative analysis of single-pulse and dual-pulse plasmas. Additionally, results indicate that temporal shaping of the nanosecond laser pulse intensity enhances dual-pulse performance relative to Gaussian pulses. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 19:00:44 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Pokharel",
"Sagar",
""
],
[
"Tropina",
"Albina",
""
]
] |
2502.00921 | Marvin Li | Marvin Li, Aayush Karan, Sitan Chen | Blink of an eye: a simple theory for feature localization in generative models | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large language models can exhibit unexpected behavior in the blink of an eye. In a recent computer use demo, a language model switched from coding to Googling pictures of Yellowstone, and these sudden shifts in behavior have also been observed in reasoning patterns and jailbreaks. This phenomenon is not unique to autoregressive models: in diffusion models, key features of the final output are decided in narrow ``critical windows'' of the generation process. In this work we develop a simple, unifying theory to explain this phenomenon using the formalism of stochastic localization samplers. We show that it emerges generically as the generation process localizes to a sub-population of the distribution it models.
While critical windows have been studied at length in diffusion models, existing theory heavily relies on strong distributional assumptions and the particulars of Gaussian diffusion. In contrast to existing work our theory (1) applies to autoregressive and diffusion models; (2) makes no distributional assumptions; (3) quantitatively improves previous bounds even when specialized to diffusions; and (4) requires basic tools and no stochastic calculus or statistical-physics-based machinery. We also identify an intriguing connection to the all-or-nothing phenomenon from statistical inference. Finally, we validate our predictions empirically for LLMs and find that critical windows often coincide with failures in problem solving for various math and reasoning benchmarks. | [
{
"version": "v1",
"created": "Sun, 02 Feb 2025 21:19:53 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 16:55:06 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"Marvin",
""
],
[
"Karan",
"Aayush",
""
],
[
"Chen",
"Sitan",
""
]
] |
2502.01110 | Palash Sarkar | Claude Carlet and Palash Sarkar | The Nonlinear Filter Model of Stream Cipher Redivivus | null | null | null | null | cs.CR | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The nonlinear filter model is an old and well understood approach to the design of secure stream ciphers. Extensive research over several decades has shown how to attack stream ciphers based on this model and has identified the security properties required of the Boolean function used as the filtering function to resist such attacks. This led to the problem of constructing Boolean functions which provide adequate security \textit{and} at the same time are efficient to implement. Unfortunately, over the last two decades no good solutions to this problem appeared in the literature. The lack of good solutions has effectively led to nonlinear filter model becoming more or less obsolete. This is a big loss to the cryptographic design toolkit, since the great advantages of the nonlinear filter model are its simplicity, well understood security and the potential to provide low cost solutions for hardware oriented stream ciphers. In this paper, we revive the nonlinear filter model by constructing appropriate Boolean functions which provide required security and are also efficient to implement. We put forward concrete suggestions of stream ciphers which are $\kappa$-bit secure against known types of attacks for $\kappa=80,128,160,192,224$ and $256$. For the $80$-bit, $128$-bit, and the $256$-bit security levels, the circuits for the corresponding stream ciphers require about 1743.5, 2771.5, and 5607.5 NAND gates respectively. For the $80$-bit and the $128$-bit security levels, the gate count estimates compare quite well to the famous ciphers Trivium and Grain-128a respectively, while for the $256$-bit security level, we do not know of any other stream cipher design which has such a low gate count. | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 07:01:21 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Feb 2025 05:37:40 GMT"
},
{
"version": "v3",
"created": "Mon, 10 Feb 2025 07:19:03 GMT"
},
{
"version": "v4",
"created": "Wed, 12 Feb 2025 10:53:03 GMT"
},
{
"version": "v5",
"created": "Tue, 29 Apr 2025 06:34:11 GMT"
},
{
"version": "v6",
"created": "Thu, 05 Jun 2025 05:50:26 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Carlet",
"Claude",
""
],
[
"Sarkar",
"Palash",
""
]
] |
2502.01575 | Tomer Meir | Tomer Meir, Uri Shalit, Malka Gorfine | Heterogeneous Treatment Effect in Time-to-Event Outcomes: Harnessing Censored Data with Recursively Imputed Trees | null | null | null | null | stat.ML cs.LG | http://creativecommons.org/licenses/by/4.0/ | Tailoring treatments to individual needs is a central goal in fields such as medicine. A key step toward this goal is estimating Heterogeneous Treatment Effects (HTE) - the way treatments impact different subgroups. While crucial, HTE estimation is challenging with survival data, where time until an event (e.g., death) is key. Existing methods often assume complete observation, an assumption violated in survival data due to right-censoring, leading to bias and inefficiency. Cui et al. (2023) proposed a doubly-robust method for HTE estimation in survival data under no hidden confounders, combining a causal survival forest with an augmented inverse-censoring weighting estimator. However, we find it struggles under heavy censoring, which is common in rare-outcome problems such as Amyotrophic lateral sclerosis (ALS). Moreover, most current methods cannot handle instrumental variables, which are a crucial tool in the causal inference arsenal. We introduce Multiple Imputation for Survival Treatment Response (MISTR), a novel, general, and non-parametric method for estimating HTE in survival data. MISTR uses recursively imputed survival trees to handle censoring without directly modeling the censoring mechanism. Through extensive simulations and analysis of two real-world datasets-the AIDS Clinical Trials Group Protocol 175 and the Illinois unemployment dataset we show that MISTR outperforms prior methods under heavy censoring in the no-hidden-confounders setting, and extends to the instrumental variable setting. To our knowledge, MISTR is the first non-parametric approach for HTE estimation with unobserved confounders via instrumental variables. | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 17:59:34 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 19:10:53 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Meir",
"Tomer",
""
],
[
"Shalit",
"Uri",
""
],
[
"Gorfine",
"Malka",
""
]
] |
2502.01760 | Patricio Escalona Contreras | Vin\'icius Oliveira, Patricio Escalona, Lucia Angel, C. A. de S. Pires, Farinaldo S. Queiroz | Type-II Seesaw Mechanism for Dirac Neutrinos and its Implications on $N_{\text{eff}}$ and Lepton Flavor Violation in a 3-3-1 model | Accepted for publication in JHEP | null | null | null | hep-ph | http://creativecommons.org/licenses/by/4.0/ | In this study, we implement the type-II seesaw mechanism for Dirac neutrino masses within the framework of a 3-3-1 model. To this end, we introduce a scalar sextet and impose both lepton number conservation and invariance under a discrete $Z_2$ symmetry in the Lagrangian. This mechanism naturally generates small Dirac neutrino masses by allowing the soft breaking of the $Z_2$ symmetry through a unique term in the scalar potential, while preserving lepton number. Specifically, we explore the realization of this model at low-energy scales. Regarding flavor implications, we analyze its contributions to the rare decay processes $\mu \to e \gamma$ and $\mu \to \bar e ee$. In the cosmological context, we analyze the influence of right-handed neutrinos on the effective number of neutrino species, $N_\text{eff}$, through interactions mediated by the $Z^{\prime}$ boson. Our findings establish a lower bound of $m_{Z^{\prime}} > 4.4$ TeV, which slightly exceeds the current lower limit set by the Large Hadron Collider (LHC). | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 19:02:37 GMT"
},
{
"version": "v2",
"created": "Mon, 10 Feb 2025 21:55:03 GMT"
},
{
"version": "v3",
"created": "Wed, 12 Feb 2025 14:15:18 GMT"
},
{
"version": "v4",
"created": "Mon, 14 Apr 2025 19:00:42 GMT"
},
{
"version": "v5",
"created": "Thu, 05 Jun 2025 05:43:35 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Oliveira",
"Vinícius",
""
],
[
"Escalona",
"Patricio",
""
],
[
"Angel",
"Lucia",
""
],
[
"Pires",
"C. A. de S.",
""
],
[
"Queiroz",
"Farinaldo S.",
""
]
] |
2502.01861 | Ethan Harvey | Ethan Harvey, Mikhail Petrov, Michael C. Hughes | Learning Hyperparameters via a Data-Emphasized Variational Objective | arXiv admin note: text overlap with arXiv:2410.19675 | null | null | null | cs.LG stat.ML | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | When training large flexible models on limited data, avoiding overfitting is a practical concern. Common grid search or smarter search methods rely on expensive separate runs at each candidate hyperparameter while carving out a validation set that reduces available training data. In this paper, we consider direct gradient-based learning of regularization hyperparameters on the full training set via the evidence lower bound ("ELBo") objective from Bayesian variational methods. We focus on scenarios where the model is over-parameterized for flexibility while the approximate posterior is chosen to be Gaussian with isotropic covariance for tractability, even though it cannot match the true posterior exactly. In such scenarios, we find the ELBo prioritizes posteriors that match the prior variance, which leads to severely underfitting the data. Instead, we recommend a data-emphasized ELBo that upweights the influence of the data likelihood relative to the prior. In Bayesian transfer learning of classifiers for text and images, our method reduces 88+ hour grid searches of past work to under 3 hours while delivering comparable accuracy. We further demonstrate how our approach enables efficient yet accurate approximations of Gaussian processes with learnable length-scale kernels. | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 22:19:35 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 03:02:11 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Harvey",
"Ethan",
""
],
[
"Petrov",
"Mikhail",
""
],
[
"Hughes",
"Michael C.",
""
]
] |
2502.01951 | Xinyi Wu | Xinyi Wu, Yifei Wang, Stefanie Jegelka, Ali Jadbabaie | On the Emergence of Position Bias in Transformers | ICML 2025 | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Recent studies have revealed various manifestations of position bias in transformer architectures, from the "lost-in-the-middle" phenomenon to attention sinks, yet a comprehensive theoretical understanding of how attention masks and positional encodings shape these biases remains elusive. This paper presents a graph-theoretic framework for analyzing position bias in multi-layer attention. Modeling attention masks as directed graphs, we quantify how tokens interact with contextual information based on their sequential positions. We uncover two key insights: First, causal masking inherently biases attention toward earlier positions, as tokens in deeper layers attend to increasingly more contextualized representations of earlier tokens. Second, we characterize the competing effects of the causal mask and relative positional encodings, such as the decay mask and rotary positional encoding (RoPE): while both mechanisms introduce distance-based decay within individual attention maps, their aggregate effect across multiple attention layers$\unicode{x2013}$coupled with the causal mask$\unicode{x2013}$leads to a trade-off between the long-term decay effects and the cumulative importance of early sequence positions. Through controlled numerical experiments, we not only validate our theoretical findings but also reproduce position biases observed in real-world LLMs. Our framework offers a principled foundation for understanding positional biases in transformers, shedding light on the complex interplay of attention mechanism components and guiding more informed architectural design. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 02:53:07 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 01:00:48 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wu",
"Xinyi",
""
],
[
"Wang",
"Yifei",
""
],
[
"Jegelka",
"Stefanie",
""
],
[
"Jadbabaie",
"Ali",
""
]
] |
2502.02014 | Jie Feng | Haohan Zou, Jie Feng, Hao Zhao, Yuanyuan Shi | Analytical Lyapunov Function Discovery: An RL-based Generative Approach | 26 pages (8+18), preprint for discussion. Haohan and Jie contribute equally | null | null | null | cs.LG cs.AI cs.SC cs.SY eess.SY | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Despite advances in learning-based methods, finding valid Lyapunov functions for nonlinear dynamical systems remains challenging. Current neural network approaches face two main issues: challenges in scalable verification and limited interpretability. To address these, we propose an end-to-end framework using transformers to construct analytical Lyapunov functions (local), which simplifies formal verification, enhances interpretability, and provides valuable insights for control engineers. Our framework consists of a transformer-based trainer that generates candidate Lyapunov functions and a falsifier that verifies candidate expressions and refines the model via risk-seeking policy gradient. Unlike Alfarano et al. (2024), which utilizes pre-training and seeks global Lyapunov functions for low-dimensional systems, our model is trained from scratch via reinforcement learning (RL) and succeeds in finding local Lyapunov functions for high-dimensional and non-polynomial systems. Given the analytical nature of the candidates, we employ efficient optimization methods for falsification during training and formal verification tools for the final verification. We demonstrate the efficiency of our approach on a range of nonlinear dynamical systems with up to ten dimensions and show that it can discover Lyapunov functions not previously identified in the control literature. Full implementation is available on \href{https://github.com/JieFeng-cse/Analytical-Lyapunov-Function-Discovery}{Github} | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 05:04:15 GMT"
},
{
"version": "v2",
"created": "Tue, 11 Feb 2025 00:19:47 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 20:16:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zou",
"Haohan",
""
],
[
"Feng",
"Jie",
""
],
[
"Zhao",
"Hao",
""
],
[
"Shi",
"Yuanyuan",
""
]
] |
2502.02939 | Hajime Kubota | Hajime Kubota | On grid homology for diagonal knots | 14 pages, 11 figures | null | null | null | math.GT | http://creativecommons.org/licenses/by/4.0/ | We partially determine grid homology (combinatorial knot Floer homology) of diagonal knots, which are conjectured to be equivalent to positive braid knots, by exploiting nice grid diagrams. We compare diagonal knots to various classes of knots, such as positive braids, fibered positive knots, and $L$-space knots. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 07:11:46 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 04:18:44 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kubota",
"Hajime",
""
]
] |
2502.03055 | Vincent Baltz | Coline Thevenard, Miina Leivisk\"a, Richard F. L. Evans, Daria Gusakova, Vincent Baltz | Imprinting of skyrmions and bimerons in an antiferromagnet | null | null | 10.1103/753s-8nx4 | null | cond-mat.mtrl-sci cond-mat.mes-hall | http://creativecommons.org/licenses/by/4.0/ | Topologically protected magnetic states in condensed matter physics, particularly antiferromagnetic (AFM) skyrmions (Sks) and bimerons (Bms), offer promising prospects for terahertz dynamics and sustained current-induced motion, thanks to their compensating multiple sub-lattice structure. However, nucleating AFM Sks and Bms is challenging due to the lack of net magnetization. Previous attempts to imprint pre-defined Sks and Bms in a ferromagnet (FM) and transfer them to an AFM using interfacial exchange bias in FM/AFM heterostructures have been hindered by complex multilayers with discontinuities, polycrystallinity, or multipartite chiral AFMs. Employing atomistic spin simulations, we demonstrate the viability of texture imprinting for nucleating Sks and Bms in AFMs, using a prototypical bipartite AFM layer in a multilayer structure free from discontinuities. Such imprinting is a crucial step towards understanding the static and dynamic properties of natural antiferromagnetic textures and their unique properties. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 10:34:51 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 13:31:31 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Thevenard",
"Coline",
""
],
[
"Leiviskä",
"Miina",
""
],
[
"Evans",
"Richard F. L.",
""
],
[
"Gusakova",
"Daria",
""
],
[
"Baltz",
"Vincent",
""
]
] |
2502.03278 | Mehdi Heydari Shahna | Mehdi Heydari Shahna, Pauli Mustalahti, and Jouni Mattila | Fault-Tolerant Control for System Availability and Continuous Operation in Heavy-Duty Wheeled Mobile Robots | This paper has been accepted by IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM2025) | null | null | null | eess.SY cs.SY | http://creativecommons.org/licenses/by/4.0/ | When the control system in a heavy-duty wheeled mobile robot (HD-WMR) malfunctions, deviations from ideal motion occur, significantly heightening the risks of off-road instability and costly damage. To meet the demands for safety, reliability, and controllability in HD-WMRs, the control system must tolerate faults to a certain extent, ensuring continuous operation. To this end, this paper introduces a model-free hierarchical control with fault accommodation (MFHCA) framework designed to address sensor and actuator faults in hydraulically powered HD-WMRs with independently controlled wheels. To begin, a novel mathematical representation of the motion dynamics of HD-WMRs, incorporating both sensor and actuator fault modes, is investigated. Subsequently, the MFHCA framework is proposed to manage all wheels under various fault modes, ensuring that each wheel tracks the reference driving velocities and steering angles, which are inverse kinematically mapped from the angular and linear velocities commanded in the HD-WMR's base frame. To do so, this framework generates appropriate power efforts in independently valve-regulated wheels to accommodate the adaptively isolated faults, thereby ensuring exponential stability. The experimental analysis of a 6,500-kg hydraulic-powered HD-WMR under various fault modes and rough terrains demonstrates the validity of the MFHCA framework. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 15:34:25 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 11:37:13 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Shahna",
"Mehdi Heydari",
""
],
[
"Mustalahti",
"Pauli",
""
],
[
"Mattila",
"Jouni",
""
]
] |
2502.03428 | Thomas Kloss | Thomas Kloss, Xavier Waintal | Propagation of ultrashort voltage pulses through a small quantum dot | 13 pages, 9 figures, published version | Phys. Rev. B 111, 235411 (2025) | 10.1103/PhysRevB.111.235411 | null | cond-mat.mes-hall | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The coherent transport of time-resolved ultrafast excitations in nanoelectronic interferometers is expected to exhibit an interesting interplay between the interferences and the time-dependent drive. However, the typical frequencies required to unlock this physics are in the THz range, making its observation challenging. In this work, we consider the propagation of the excitation generated by ultrashort voltage pulses through a small quantum dot, a system which we argue can display similar physics at significantly lower frequencies. We model the system with a single resonant level connected to two infinite electrodes subjected to a time-dependent voltage bias. For short pulses, we predict that the behaviour of the dot contrasts sharply with the long pulse (adiabatic) limit: the current actually oscillates with the amplitude of the voltage pulse. In the ultrafast limit, we predict that the current can even be negative, i.e. flow against the voltage drop. Our results are obtained by a combination of two approaches that are in quantitative agreement: explicit analytical expressions in the ultrafast and ultraslow limits and exact numerical simulations. We discuss the applicability of our findings and conclude that this system should be within reach of existing experimental platforms. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 18:18:09 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 06:54:15 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kloss",
"Thomas",
""
],
[
"Waintal",
"Xavier",
""
]
] |
2502.03740 | Hee-Jun Jung | Hee-Jun Jung, Jaehyoung Jeong, Kangil Kim | Multiple Invertible and Partial-Equivariant Function for Latent Vector Transformation to Enhance Disentanglement in VAEs | This work has been submitted to the IEEE for possible publication | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Disentanglement learning is a core issue for understanding and re-using trained information in Variational AutoEncoder (VAE), and effective inductive bias has been reported as a key factor. However, the actual implementation of such bias is still vague. In this paper, we propose a novel method, called Multiple Invertible and partial-equivariant transformation (MIPE-transformation), to inject inductive bias by 1) guaranteeing the invertibility of latent-to-latent vector transformation while preserving a certain portion of equivariance of input-to-latent vector transformation, called Invertible and partial-equivariant transformation (IPE-transformation), 2) extending the form of prior and posterior in VAE frameworks to an unrestricted form through a learnable conversion to an approximated exponential family, called Exponential Family conversion (EF-conversion), and 3) integrating multiple units of IPE-transformation and EF-conversion, and their training. In experiments on 3D Cars, 3D Shapes, and dSprites datasets, MIPE-transformation improves the disentanglement performance of state-of-the-art VAEs. | [
{
"version": "v1",
"created": "Thu, 06 Feb 2025 03:08:12 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 06:28:20 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jung",
"Hee-Jun",
""
],
[
"Jeong",
"Jaehyoung",
""
],
[
"Kim",
"Kangil",
""
]
] |
2502.04507 | Peiyuan Zhang | Peiyuan Zhang, Yongqi Chen, Runlong Su, Hangliang Ding, Ion Stoica, Zhengzhong Liu, Hao Zhang | Fast Video Generation with Sliding Tile Attention | Accepted by ICML 2025 | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Diffusion Transformers (DiTs) with 3D full attention power state-of-the-art video generation, but suffer from prohibitive compute cost -- when generating just a 5-second 720P video, attention alone takes 800 out of 945 seconds of total inference time. This paper introduces sliding tile attention (STA) to address this challenge. STA leverages the observation that attention scores in pretrained video diffusion models predominantly concentrate within localized 3D windows. By sliding and attending over the local spatial-temporal region, STA eliminates redundancy from full attention. Unlike traditional token-wise sliding window attention (SWA), STA operates tile-by-tile with a novel hardware-aware sliding window design, preserving expressiveness while being hardware-efficient. With careful kernel-level optimizations, STA offers the first efficient 2D/3D sliding-window-like attention implementation, achieving 58.79% MFU. Precisely, STA accelerates attention by 2.8-17x over FlashAttention-2 (FA2) and 1.6-10x over FlashAttention-3 (FA3). On the leading video DiT, HunyuanVideo, STA reduces end-to-end latency from 945s (FA3) to 685s without quality degradation, requiring no training. Enabling finetuning further lowers latency to 268s with only a 0.09% drop on VBench. We make our codebase public at https://github.com/hao-ai-lab/FastVideo. | [
{
"version": "v1",
"created": "Thu, 06 Feb 2025 21:17:09 GMT"
},
{
"version": "v2",
"created": "Mon, 26 May 2025 04:37:29 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 23:21:39 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Peiyuan",
""
],
[
"Chen",
"Yongqi",
""
],
[
"Su",
"Runlong",
""
],
[
"Ding",
"Hangliang",
""
],
[
"Stoica",
"Ion",
""
],
[
"Liu",
"Zhengzhong",
""
],
[
"Zhang",
"Hao",
""
]
] |
2502.05221 | Jun Pyo Seo | Jun Pyo Seo | Blackout DIFUSCO | 12 pages | null | null | null | math.OC cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This study explores the integration of Blackout Diffusion into the DIFUSCO framework for combinatorial optimization, specifically targeting the Traveling Salesman Problem (TSP). Inspired by the success of discrete-time diffusion models (D3PM) in maintaining structural integrity, we extend the paradigm to a continuous-time framework, leveraging the unique properties of Blackout Diffusion. Continuous-time modeling introduces smoother transitions and refined control, hypothesizing enhanced solution quality over traditional discrete methods. We propose three key improvements to enhance the diffusion process. First, we transition from a discrete-time-based model to a continuous-time framework, providing a more refined and flexible formulation. Second, we refine the observation time scheduling to ensure a smooth and linear transformation throughout the diffusion process, allowing for a more natural progression of states. Finally, building upon the second improvement, we further enhance the reverse process by introducing finer time slices in regions that are particularly challenging for the model, thereby improving accuracy and stability in the reconstruction phase. Although the experimental results did not exceed the baseline performance, they demonstrate the effectiveness of these methods in balancing simplicity and complexity, offering new insights into diffusion-based combinatorial optimization. This work represents the first application of Blackout Diffusion to combinatorial optimization, providing a foundation for further advancements in this domain. * The code is available for review at https://github.com/Giventicket/BlackoutDIFUSCO. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 17:24:33 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 09:17:14 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Seo",
"Jun Pyo",
""
]
] |
2502.05625 | Jacob Christopher | Stefano Zampini, Jacob Christopher, Luca Oneto, Davide Anguita, Ferdinando Fioretto | Training-Free Constrained Generation With Stable Diffusion Models | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Stable diffusion models represent the state-of-the-art in data synthesis across diverse domains and hold transformative potential for applications in science and engineering, e.g., by facilitating the discovery of novel solutions and simulating systems that are computationally intractable to model explicitly. While there is increasing effort to incorporate physics-based constraints into generative models, existing techniques are either limited in their applicability to latent diffusion frameworks or lack the capability to strictly enforce domain-specific constraints. To address this limitation this paper proposes a novel integration of stable diffusion models with constrained optimization frameworks, enabling the generation of outputs satisfying stringent physical and functional requirements. The effectiveness of this approach is demonstrated through material design experiments requiring adherence to precise morphometric properties, challenging inverse design tasks involving the generation of materials inducing specific stress-strain responses, and copyright-constrained content generation tasks. | [
{
"version": "v1",
"created": "Sat, 08 Feb 2025 16:11:17 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 18:24:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zampini",
"Stefano",
""
],
[
"Christopher",
"Jacob",
""
],
[
"Oneto",
"Luca",
""
],
[
"Anguita",
"Davide",
""
],
[
"Fioretto",
"Ferdinando",
""
]
] |
2502.05835 | Cuipeng Wang | Cuipeng Wang, Tieyuan Chen, Haipeng Wang | Contrastive Representation Distillation via Multi-Scale Feature Decoupling | null | null | null | null | cs.CV cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Knowledge distillation is a technique aimed at enhancing the performance of a small student network without increasing its parameter size by transferring knowledge from a large, pre-trained teacher network. In the feature space, different local regions within an individual global feature map often encode distinct yet interdependent semantic information. However, previous methods mainly focus on transferring global feature knowledge, neglecting the decoupling of interdependent local regions within an individual global feature, which often results in suboptimal performance. To address this limitation, we propose MSDCRD, a novel contrastive representation distillation approach that explicitly performs multi-scale decoupling within the feature space. MSDCRD employs a multi-scale sliding-window pooling approach within the feature space to capture representations at various granularities effectively. This, in conjunction with sample categorization, facilitates efficient multi-scale feature decoupling. When integrated with a novel and effective contrastive loss function, this forms the core of MSDCRD. Feature representations differ significantly across network architectures, and this divergence becomes more pronounced in heterogeneous models, rendering feature distillation particularly challenging. Despite this, our method not only achieves superior performance in homogeneous models but also enables efficient feature knowledge transfer across a variety of heterogeneous teacher-student pairs, highlighting its strong generalizability. Moreover, its plug-and-play and parameter-free nature enables flexible integration with different visual tasks. Extensive experiments on different visual benchmarks consistently confirm the superiority of our method in enhancing the performance of student models. | [
{
"version": "v1",
"created": "Sun, 09 Feb 2025 10:03:18 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 03:15:28 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wang",
"Cuipeng",
""
],
[
"Chen",
"Tieyuan",
""
],
[
"Wang",
"Haipeng",
""
]
] |
2502.05954 | Amal Aldawsari | Amal Aldawsari and Evangelos Pournaras | Optimization under Attack: Resilience, Vulnerability, and the Path to Collapse | null | null | null | null | cs.MA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Optimization is instrumental for improving operations of large-scale socio-technical infrastructures of Smart Cities, for instance, energy and traffic systems. In particular, understanding the performance of multi-agent discrete-choice combinatorial optimization under distributed adversary attacks is a compelling and underexplored problem, since multi-agent systems exhibit a large number of remote control variables that can influence in an unprecedented way the cost-effectiveness of distributed optimization heuristics. This paper unravels for the first time the trajectories of distributed optimization from resilience to vulnerability, and finally to collapse under varying adversary influence. Using real-world data to emulate over 28 billion multi-agent optimization scenarios, we exhaustively assess how the number of agents with different adversarial severity and network positioning influences optimization performance, including the influence on Pareto optimal points. With this novel large-scale dataset, made openly available as a benchmark, we disentangle how optimization remains resilient to adversaries and which adversary conditions are required to make optimization vulnerable or collapsed. These new findings can provide new insights for designing self-healing strategies for fault-tolerance and fault-correction in adversarial distributed optimization that have been missing so far. | [
{
"version": "v1",
"created": "Sun, 09 Feb 2025 16:48:09 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 21:56:24 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Aldawsari",
"Amal",
""
],
[
"Pournaras",
"Evangelos",
""
]
] |
2502.05967 | Abhay Gupta | Saaketh Narayan, Abhay Gupta, Mansheej Paul, Davis Blalock | $\mu$nit Scaling: Simple and Scalable FP8 LLM Training | ICML 2025; 17 pages, 13 figures | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large Language Model training with 8-bit floating point (FP8) formats promises significant efficiency improvements, but reduced numerical precision makes training challenging. It is currently possible to train in FP8 only if one is willing to tune various hyperparameters, reduce model scale, or accept the overhead of computing dynamic scale factors. We demonstrate simple, scalable FP8 training that requires no dynamic scaling factors or special hyperparameters, even at large model sizes. Our method, $\mu$nit Scaling ($\mu$S), also enables simple hyperparameter transfer across model widths, matched numerics across training and inference, and other desirable properties. $\mu$nit Scaling is straightforward to implement, consisting of a set of minimal interventions based on a first-principles analysis of common transformer operations. We validate our method by training models from 1B to 13B parameters, performing all hidden linear layer computations in FP8. We achieve quality equal to higher precision baselines while also training up to 33% faster. | [
{
"version": "v1",
"created": "Sun, 09 Feb 2025 17:31:09 GMT"
},
{
"version": "v2",
"created": "Sun, 02 Mar 2025 20:16:43 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 08:10:47 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Narayan",
"Saaketh",
""
],
[
"Gupta",
"Abhay",
""
],
[
"Paul",
"Mansheej",
""
],
[
"Blalock",
"Davis",
""
]
] |
2502.06072 | Yige Hong | Xiangcheng Zhang, Yige Hong, Weina Wang | Projection-based Lyapunov method for fully heterogeneous weakly-coupled MDPs | 34 pages, updated related work to include a missing result | null | null | null | cs.LG math.OC math.PR | http://creativecommons.org/licenses/by/4.0/ | Heterogeneity poses a fundamental challenge for many real-world large-scale decision-making problems but remains largely understudied. In this paper, we study the fully heterogeneous setting of a prominent class of such problems, known as weakly-coupled Markov decision processes (WCMDPs). Each WCMDP consists of $N$ arms (or subproblems), which have distinct model parameters in the fully heterogeneous setting, leading to the curse of dimensionality when $N$ is large. We show that, under mild assumptions, an efficiently computable policy achieves an $O(1/\sqrt{N})$ optimality gap in the long-run average reward per arm for fully heterogeneous WCMDPs as $N$ becomes large. This is the first asymptotic optimality result for fully heterogeneous average-reward WCMDPs. Our main technical innovation is the construction of projection-based Lyapunov functions that certify the convergence of rewards and costs to an optimal region, even under full heterogeneity. | [
{
"version": "v1",
"created": "Sun, 09 Feb 2025 23:32:19 GMT"
},
{
"version": "v2",
"created": "Thu, 20 Feb 2025 04:25:14 GMT"
},
{
"version": "v3",
"created": "Mon, 19 May 2025 05:28:13 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 01:04:57 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Xiangcheng",
""
],
[
"Hong",
"Yige",
""
],
[
"Wang",
"Weina",
""
]
] |
2502.06466 | Ahmad Rafsanjani | Burcu Seyido\u{g}lu, Aida Parvaresh, Bahman Taherkhani, Ahmad Rafsanjani | Inflatable Kirigami Crawlers | null | null | null | null | cs.RO | http://creativecommons.org/licenses/by/4.0/ | Kirigami offers unique opportunities for guided morphing by leveraging the geometry of the cuts. This work presents inflatable kirigami crawlers created by introducing cut patterns into heat-sealable textiles to achieve locomotion upon cyclic pneumatic actuation. Inflating traditional air pouches results in symmetric bulging and contraction. In inflated kirigami actuators, the accumulated compressive forces uniformly break the symmetry, enhance contraction compared to simple air pouches by two folds, and trigger local rotation of the sealed edges that overlap and self-assemble into an architected surface with emerging scale-like features. As a result, the inflatable kirigami actuators exhibit a uniform, controlled contraction with asymmetric localized out-of-plane deformations. This process allows us to harness the geometric and material nonlinearities to imbue inflatable textile-based kirigami actuators with predictable locomotive functionalities. We thoroughly characterized the programmed deformations of these actuators and their impact on friction. We found that the kirigami actuators exhibit directional anisotropic friction properties when inflated, having higher friction coefficients against the direction of the movement, enabling them to move across surfaces with varying roughness. We further enhanced the functionality of inflatable kirigami actuators by introducing multiple channels and segments to create functional soft robotic prototypes with versatile locomotion capabilities. | [
{
"version": "v1",
"created": "Mon, 10 Feb 2025 13:47:10 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 01:31:06 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Seyidoğlu",
"Burcu",
""
],
[
"Parvaresh",
"Aida",
""
],
[
"Taherkhani",
"Bahman",
""
],
[
"Rafsanjani",
"Ahmad",
""
]
] |
2502.06475 | Andrew Cleary | Andrew Cleary and Jacob Page | Dynamical relevance of periodic orbits under increasing Reynolds number and connections to inviscid dynamics | null | null | null | null | physics.flu-dyn | http://creativecommons.org/licenses/by/4.0/ | Large numbers of relative periodic orbits (RPOs) have been found recently in doubly-periodic, two-dimensional Kolmogorov flow at moderate Reynolds numbers $Re \in \{40, 100\}$. While these solutions lead to robust statistical reconstructions at the $Re$-values where they were obtained, it is unclear how their dynamical importance evolves with increasing $Re$. We perform arclength continuation on this library of solutions to show that large numbers of RPOs quickly become dynamically irrelevant, reaching dissipation values either well above or below those associated with the turbulent attractor at high $Re$. The scaling of the high dissipation RPOs is shown to be consistent with a direct connection to solutions of the unforced Euler equation, and is observed for a wide variety of states beyond the 'unimodal' solutions considered in previous work (Kim & Okamoto, Nonlinearity 28, 2015). On the other hand, the weakly dissipative states have properties indicating a connection to exact solutions of a forced Euler equation. The apparent dynamical irrelevance is associated with poor statistical reconstructions away from the $Re$ values where the RPOs were originally converged. Motivated by the connection to solutions of the Euler equation, we show that many of these states can be well described by exact relative periodic solutions in a system of point vortices. The point vortex RPOs are converged via gradient-based optimisation of a scalar loss function which (1) matches the dynamics of the point vortices to the turbulent vortex cores and (2) insists the point vortex evolution is itself time-periodic. | [
{
"version": "v1",
"created": "Mon, 10 Feb 2025 13:54:53 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:55:40 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Cleary",
"Andrew",
""
],
[
"Page",
"Jacob",
""
]
] |
2502.06765 | Manuel M\"uller | Manuel M. M\"uller, Yuetian Luo, Rina Foygel Barber | Are all models wrong? Fundamental limits in distribution-free empirical model falsification | 39 pages, 1 figure | null | null | null | math.ST cs.LG stat.ML stat.TH | http://creativecommons.org/licenses/by/4.0/ | In statistics and machine learning, when we train a fitted model on available data, we typically want to ensure that we are searching within a model class that contains at least one accurate model -- that is, we would like to ensure an upper bound on the model class risk (the lowest possible risk that can be attained by any model in the class). However, it is also of interest to establish lower bounds on the model class risk, for instance so that we can determine whether our fitted model is at least approximately optimal within the class, or, so that we can decide whether the model class is unsuitable for the particular task at hand. Particularly in the setting of interpolation learning where machine learning models are trained to reach zero error on the training data, we might ask if, at the very least, a positive lower bound on the model class risk is possible -- or are we unable to detect that "all models are wrong"? In this work, we answer these questions in a distribution-free setting by establishing a model-agnostic, fundamental hardness result for the problem of constructing a lower bound on the best test error achievable over a model class, and examine its implications on specific model classes such as tree-based methods and linear regression. | [
{
"version": "v1",
"created": "Mon, 10 Feb 2025 18:44:30 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 11:34:12 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Müller",
"Manuel M.",
""
],
[
"Luo",
"Yuetian",
""
],
[
"Barber",
"Rina Foygel",
""
]
] |
2502.06854 | Hailong Jiang | Hailong Jiang, Jianfeng Zhu, Yao Wan, Bo Fang, Hongyu Zhang, Ruoming Jin, Qiang Guan | Can Large Language Models Understand Intermediate Representations in Compilers? | null | null | null | null | cs.LG cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Intermediate Representations (IRs) play a critical role in compiler design and program analysis, yet their comprehension by Large Language Models (LLMs) remains underexplored. In this paper, we present an explorative empirical study evaluating the capabilities of six state-of-the-art LLMs: GPT-4, GPT-3, DeepSeek, Gemma 2, Llama 3, and Code Llama, in understanding IRs. Specifically, we assess model performance across four core tasks: control flow graph reconstruction, decompilation, code summarization, and execution reasoning. While LLMs exhibit competence in parsing IR syntax and identifying high-level structures, they consistently struggle with instruction-level reasoning, especially in control flow reasoning, loop handling, and dynamic execution. Common failure modes include misinterpreting branching instructions, omitting critical operations, and relying on heuristic reasoning rather than precise instruction-level logic. Our findings highlight the need for IR-specific enhancements in LLM design. We recommend fine-tuning on structured IR datasets and integrating control-flow-sensitive architectures to improve model effectiveness. All experimental data and source code are publicly available at | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 17:23:48 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:48:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jiang",
"Hailong",
""
],
[
"Zhu",
"Jianfeng",
""
],
[
"Wan",
"Yao",
""
],
[
"Fang",
"Bo",
""
],
[
"Zhang",
"Hongyu",
""
],
[
"Jin",
"Ruoming",
""
],
[
"Guan",
"Qiang",
""
]
] |
2502.07987 | Temurbek Rahmatullaev | Temurbek Rahmatullaev, Polina Druzhinina, Nikita Kurdiukov, Matvey Mikhalchuk, Andrey Kuznetsov, Anton Razzhigaev | Universal Adversarial Attack on Aligned Multimodal LLMs | Added benchmarks, baselines, author, appendix | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | We propose a universal adversarial attack on multimodal Large Language Models (LLMs) that leverages a single optimized image to override alignment safeguards across diverse queries and even multiple models. By backpropagating through the vision encoder and language head, we craft a synthetic image that forces the model to respond with a targeted phrase (e.g., "Sure, here it is") or otherwise unsafe content -- even for harmful prompts. In experiments on the SafeBench and MM-SafetyBench benchmarks, our method achieves higher attack success rates than existing baselines, including text-only universal prompts (e.g., up to 81% on certain models). We further demonstrate cross-model universality by training on several multimodal LLMs simultaneously. Additionally, a multi-answer variant of our approach produces more natural-sounding (yet still malicious) responses. These findings underscore critical vulnerabilities in current multimodal alignment and call for more robust adversarial defenses. We will release code and datasets under the Apache-2.0 license. Warning: some content generated by Multimodal LLMs in this paper may be offensive. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 22:07:47 GMT"
},
{
"version": "v2",
"created": "Thu, 13 Feb 2025 06:40:14 GMT"
},
{
"version": "v3",
"created": "Wed, 04 Jun 2025 18:10:51 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Rahmatullaev",
"Temurbek",
""
],
[
"Druzhinina",
"Polina",
""
],
[
"Kurdiukov",
"Nikita",
""
],
[
"Mikhalchuk",
"Matvey",
""
],
[
"Kuznetsov",
"Andrey",
""
],
[
"Razzhigaev",
"Anton",
""
]
] |
2502.08094 | Dr. Goutam Manna | Arijit Panda and Eduardo Guendelman and Goutam Manna | Exploring Cosmological Implications of the Modified Raychaudhuri Equation in Non-Gravitating Vacuum Energy Theory | 20 pages, 10 figures, Comments are welcome | Eur. Phys. J. C 85, 581 (2025) | 10.1140/epjc/s10052-025-14309-4 | null | gr-qc hep-th | http://creativecommons.org/licenses/by-sa/4.0/ | This article investigates the modified Raychaudhuri Equation (RE) in the context of Non-Gravitating Vacuum Energy (NGVE) theory and its implications for various cosmological characteristics. The equation is formulated based on the NGVE framework, in which global scale invariance generates a unique geometry. The newly developed geometry introduces a metric that is conformally connected to the conventional metric, with the conformal factor dependent on scalar field potentials. The cosmological study is carried out under the framework of a flat Friedmann-Lema\^itre-Robertson-Walker (FLRW) universe. Assuming matter behaves as an ideal fluid in the modified geometry, we formulate models for conditional expansion, collapse, and steady state, governed by the scalar field ($\phi$). In this context, the caustic solution and the focusing theorem are also studied. Scalar field solutions for exponential and power-law scale factors are also derived using NGVE theory's equations of motion. Finally, graphical analysis is used to investigate the behavior of the interaction terms that appear in the modified RE under these scale factors. | [
{
"version": "v1",
"created": "Wed, 12 Feb 2025 03:45:43 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Panda",
"Arijit",
""
],
[
"Guendelman",
"Eduardo",
""
],
[
"Manna",
"Goutam",
""
]
] |
2502.08503 | Yanheng He | Jiahe Jin, Yanheng He, Mingyan Yang | Revisiting 3D LLM Benchmarks: Are We Really Testing 3D Capabilities? | Accepted to ACL 2025 Findings | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this work, we identify the "2D-Cheating" problem in 3D LLM evaluation, where these tasks might be easily solved by VLMs with rendered images of point clouds, exposing ineffective evaluation of 3D LLMs' unique 3D capabilities. We test VLM performance across multiple 3D LLM benchmarks and, using this as a reference, propose principles for better assessing genuine 3D understanding. We also advocate explicitly separating 3D abilities from 1D or 2D aspects when evaluating 3D LLMs. Code and data are available at https://github.com/LLM-class-group/Revisiting-3D-LLM-Benchmarks . | [
{
"version": "v1",
"created": "Wed, 12 Feb 2025 15:34:45 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 16:35:34 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jin",
"Jiahe",
""
],
[
"He",
"Yanheng",
""
],
[
"Yang",
"Mingyan",
""
]
] |
2502.08605 | Karish Grover | Karish Grover, Geoffrey J. Gordon, Christos Faloutsos | CurvGAD: Leveraging Curvature for Enhanced Graph Anomaly Detection | ICML 2025 | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Does the intrinsic curvature of complex networks hold the key to unveiling graph anomalies that conventional approaches overlook? Reconstruction-based graph anomaly detection (GAD) methods overlook such geometric outliers, focusing only on structural and attribute-level anomalies. To this end, we propose CurvGAD - a mixed-curvature graph autoencoder that introduces the notion of curvature-based geometric anomalies. CurvGAD introduces two parallel pipelines for enhanced anomaly interpretability: (1) Curvature-equivariant geometry reconstruction, which focuses exclusively on reconstructing the edge curvatures using a mixed-curvature, Riemannian encoder and Gaussian kernel-based decoder; and (2) Curvature-invariant structure and attribute reconstruction, which decouples structural and attribute anomalies from geometric irregularities by regularizing graph curvature under discrete Ollivier-Ricci flow, thereby isolating the non-geometric anomalies. By leveraging curvature, CurvGAD refines the existing anomaly classifications and identifies new curvature-driven anomalies. Extensive experimentation over 10 real-world datasets (both homophilic and heterophilic) demonstrates an improvement of up to 6.5% over state-of-the-art GAD methods. The code is available at: https://github.com/karish-grover/curvgad. | [
{
"version": "v1",
"created": "Wed, 12 Feb 2025 17:49:46 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 20:09:51 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Grover",
"Karish",
""
],
[
"Gordon",
"Geoffrey J.",
""
],
[
"Faloutsos",
"Christos",
""
]
] |
2502.08852 | Sarah Strikwerda | Sarah Strikwerda, Hung Vinh Tran, and Minh-Binh Tran | Controlling discrete semilinear wave equations toward flocks | null | null | null | null | math.OC | http://creativecommons.org/licenses/by-nc-nd/4.0/ | In this work, we initiate the research on controlling nonlinear waves propagating on lattices from a completely new perspective. We consider nonlinear waves on a lattice as a system of interacting particles and study their collective flocking behavior. By designing suitable feedback controls, we show that any admissible flock can be reached within a finite amount of time. Finally, we highlight the connection between our flocking problem and a minimal-time problem in the framework of nonlinear Hamilton-Jacobi equations and optimal control theory. | [
{
"version": "v1",
"created": "Wed, 12 Feb 2025 23:54:10 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 19:49:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Strikwerda",
"Sarah",
""
],
[
"Tran",
"Hung Vinh",
""
],
[
"Tran",
"Minh-Binh",
""
]
] |
2502.09080 | Wang Qiwei | Qiwei Wang, Shaoxun Wu, Yujiao Shi | BevSplat: Resolving Height Ambiguity via Feature-Based Gaussian Primitives for Weakly-Supervised Cross-View Localization | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | This paper addresses the problem of weakly supervised cross-view localization, where the goal is to estimate the pose of a ground camera relative to a satellite image with noisy ground truth annotations. A common approach to bridge the cross-view domain gap for pose estimation is Bird's-Eye View (BEV) synthesis. However, existing methods struggle with height ambiguity due to the lack of depth information in ground images and satellite height maps. Previous solutions either assume a flat ground plane or rely on complex models, such as cross-view transformers. We propose BevSplat, a novel method that resolves height ambiguity by using feature-based Gaussian primitives. Each pixel in the ground image is represented by a 3D Gaussian with semantic and spatial features, which are synthesized into a BEV feature map for relative pose estimation. Additionally, to address challenges with panoramic query images, we introduce an icosphere-based supervision strategy for the Gaussian primitives. We validate our method on the widely used KITTI and VIGOR datasets, which include both pinhole and panoramic query images. Experimental results show that BevSplat significantly improves localization accuracy over prior approaches. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 08:54:04 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:28:06 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wang",
"Qiwei",
""
],
[
"Wu",
"Shaoxun",
""
],
[
"Shi",
"Yujiao",
""
]
] |
2502.09257 | Liad Erez | Liad Erez and Tomer Koren | Bandit Multiclass List Classification | null | null | null | null | cs.LG cs.AI stat.ML | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study the problem of multiclass list classification with (semi-)bandit feedback, where input examples are mapped into subsets of size $m$ of a collection of $K$ possible labels. In each round of the interaction, the learner observes feedback consisting of the predicted labels which lie in some underlying set of ground truth labels associated with the given example. Our main result is for the $(\varepsilon,\delta)$-PAC variant of the problem for which we design an algorithm that returns an $\varepsilon$-optimal hypothesis with high probability using a sample complexity of $\widetilde{O} \big( (\mathrm{poly}(K/m) + sm / \varepsilon^2) \log (|H|/\delta) \big)$ where $H$ is the underlying (finite) hypothesis class and $s$ is an upper bound on the number of true labels for a given example. This bound improves upon known bounds for combinatorial semi-bandits whenever $s \ll K$. Moreover, in the regime where $s = O(1)$ the leading terms in our bound match the corresponding full-information rates, implying that bandit feedback essentially comes at no cost. Our PAC learning algorithm is also computationally efficient given access to an ERM oracle for $H$. In the special case of single-label classification corresponding to $s=m=1$, we prove a sample complexity bound of $O \big((K^7 + 1/\varepsilon^2)\log (|H|/\delta)\big)$ which improves upon recent results in this scenario (Erez et al. '24). Additionally, we consider the regret minimization setting where data can be generated adversarially, and establish a regret bound of $\widetilde O(|H| + \sqrt{smT \log |H|})$. Our results generalize and extend prior work in the simpler single-label setting (Erez et al. '24), and apply more generally to contextual combinatorial semi-bandit problems with $s$-sparse rewards. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 12:13:25 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 07:39:15 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Erez",
"Liad",
""
],
[
"Koren",
"Tomer",
""
]
] |
2502.09560 | Rui Yang | Rui Yang, Hanyang Chen, Junyu Zhang, Mark Zhao, Cheng Qian, Kangrui Wang, Qineng Wang, Teja Venkat Koripella, Marziyeh Movahedi, Manling Li, Heng Ji, Huan Zhang, Tong Zhang | EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents | Accepted to ICML 2025 | null | null | null | cs.AI cs.CL cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Leveraging Multi-modal Large Language Models (MLLMs) to create embodied agents offers a promising avenue for tackling real-world tasks. While language-centric embodied agents have garnered substantial attention, MLLM-based embodied agents remain underexplored due to the lack of comprehensive evaluation frameworks. To bridge this gap, we introduce EmbodiedBench, an extensive benchmark designed to evaluate vision-driven embodied agents. EmbodiedBench features: (1) a diverse set of 1,128 testing tasks across four environments, ranging from high-level semantic tasks (e.g., household) to low-level tasks involving atomic actions (e.g., navigation and manipulation); and (2) six meticulously curated subsets evaluating essential agent capabilities like commonsense reasoning, complex instruction understanding, spatial awareness, visual perception, and long-term planning. Through extensive experiments, we evaluated 24 leading proprietary and open-source MLLMs within EmbodiedBench. Our findings reveal that: MLLMs excel at high-level tasks but struggle with low-level manipulation, with the best model, GPT-4o, scoring only 28.9\% on average. EmbodiedBench provides a multifaceted standardized evaluation platform that not only highlights existing challenges but also offers valuable insights to advance MLLM-based embodied agents. Our code and dataset are available at https://embodiedbench.github.io. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 18:11:34 GMT"
},
{
"version": "v2",
"created": "Sun, 23 Feb 2025 07:30:59 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 07:22:50 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yang",
"Rui",
""
],
[
"Chen",
"Hanyang",
""
],
[
"Zhang",
"Junyu",
""
],
[
"Zhao",
"Mark",
""
],
[
"Qian",
"Cheng",
""
],
[
"Wang",
"Kangrui",
""
],
[
"Wang",
"Qineng",
""
],
[
"Koripella",
"Teja Venkat",
""
],
[
"Movahedi",
"Marziyeh",
""
],
[
"Li",
"Manling",
""
],
[
"Ji",
"Heng",
""
],
[
"Zhang",
"Huan",
""
],
[
"Zhang",
"Tong",
""
]
] |
2502.09755 | Chaim Baskin | Amit Levi, Rom Himelstein, Yaniv Nemcovsky, Avi Mendelson, Chaim Baskin | Jailbreak Attack Initializations as Extractors of Compliance Directions | null | null | null | null | cs.CR cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Safety-aligned LLMs respond to prompts with either compliance or refusal, each corresponding to distinct directions in the model's activation space. Recent works show that initializing attacks via self-transfer from other prompts significantly enhances their performance. However, the underlying mechanisms of these initializations remain unclear, and attacks utilize arbitrary or hand-picked initializations. This work presents that each gradient-based jailbreak attack and subsequent initialization gradually converge to a single compliance direction that suppresses refusal, thereby enabling an efficient transition from refusal to compliance. Based on this insight, we propose CRI, an initialization framework that aims to project unseen prompts further along compliance directions. We demonstrate our approach on multiple attacks, models, and datasets, achieving an increased attack success rate (ASR) and reduced computational overhead, highlighting the fragility of safety-aligned LLMs. A reference implementation is available at: https://amit1221levi.github.io/CRI-Jailbreak-Init-LLMs-evaluation. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 20:25:40 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 15:08:36 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Levi",
"Amit",
""
],
[
"Himelstein",
"Rom",
""
],
[
"Nemcovsky",
"Yaniv",
""
],
[
"Mendelson",
"Avi",
""
],
[
"Baskin",
"Chaim",
""
]
] |
2502.10158 | Joongkyu Lee | Joongkyu Lee, Min-hwan Oh | Combinatorial Reinforcement Learning with Preference Feedback | Accepted at ICML 2025 | null | null | null | stat.ML cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | In this paper, we consider combinatorial reinforcement learning with preference feedback, where a learning agent sequentially offers an action--an assortment of multiple items to--a user, whose preference feedback follows a multinomial logistic (MNL) model. This framework allows us to model real-world scenarios, particularly those involving long-term user engagement, such as in recommender systems and online advertising. However, this framework faces two main challenges: (1) the unknown value of each item, unlike traditional MNL bandits that only address single-step preference feedback, and (2) the difficulty of ensuring optimism while maintaining tractable assortment selection in the combinatorial action space with unknown values. In this paper, we assume a contextual MNL preference model, where the mean utilities are linear, and the value of each item is approximated by a general function. We propose an algorithm, MNL-VQL, that addresses these challenges, making it both computationally and statistically efficient. As a special case, for linear MDPs (with the MNL preference feedback), we establish the first regret lower bound in this framework and show that MNL-VQL achieves nearly minimax-optimal regret. To the best of our knowledge, this is the first work to provide statistical guarantees in combinatorial RL with preference feedback. | [
{
"version": "v1",
"created": "Fri, 14 Feb 2025 13:37:02 GMT"
},
{
"version": "v2",
"created": "Wed, 28 May 2025 02:29:42 GMT"
},
{
"version": "v3",
"created": "Thu, 05 Jun 2025 01:30:34 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Lee",
"Joongkyu",
""
],
[
"Oh",
"Min-hwan",
""
]
] |
2502.10613 | Ben Adcock | Ben Adcock | Function recovery and optimal sampling in the presence of nonuniform evaluation costs | null | null | null | null | math.NA cs.NA | http://creativecommons.org/licenses/by/4.0/ | We consider recovering a function $f : D \rightarrow \mathbb{C}$ in an $n$-dimensional linear subspace $\mathcal{P}$ from i.i.d. pointwise samples via (weighted) least-squares estimators. Different from most works, we assume the cost of evaluating $f$ is potentially nonuniform, and governed by a cost function $c : D \rightarrow (0,\infty)$ which may blow up at certain points. We therefore strive to choose the sampling measure in a way that minimizes the expected total cost. We provide a recovery guarantee which asserts accurate and stable recovery with an expected cost depending on the Christoffel function and Remez constant of the space $\mathcal{P}$. This leads to a general recipe for finding a good sampling measure for general $c$. As an example, we consider one-dimensional polynomial spaces. Here, we provide two strategies for choosing the sampling measure, which we prove are optimal (up to constants and log factors) in the case of algebraically-growing cost functions. | [
{
"version": "v1",
"created": "Sat, 15 Feb 2025 00:19:59 GMT"
},
{
"version": "v2",
"created": "Wed, 04 Jun 2025 19:49:16 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Adcock",
"Ben",
""
]
] |
2502.10797 | Dibakar Roychowdhury | Dibakar Roychowdhury | Holographic timelike entanglement and $c$ theorem for supersymmetric QFTs in ($ 0+1 $)d | v2, added figure and a footnote, Accepted in JHEP | JHEP 06 (2025) 003 | 10.1007/JHEP06(2025)003 | null | hep-th gr-qc | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present a holographic set up that computes timelike Entanglement Entropy (tEE) in $ (0+1) $d QFTs preserving some amount of SUSY. The first example we consider is that of $\mathcal{N}=2$ matrix models with massive deformations. These are dual to non-Abelian T-dual of $AdS_5 \times S^5$ that asymptotes to \emph{smeared} D0 branes. The second example, that we consider is of $ \mathcal{N}=4 $ superconformal quantum mechanical quivers in ($ 0+1 $)d that are dual to a class of type IIB backgrounds with an $ AdS_2 $ factor. In both of these examples, tEE reveals a remarkable similarity with holographic $ c $ function pertaining to a RG flow. We further compute the complexity in these models, which also reveals an identical behaviour indicating the fact that tEE is a measure of number of degrees of freedom for these ($ 0+1 $)d SQFTs in a RG flow from UV to deep IR. | [
{
"version": "v1",
"created": "Sat, 15 Feb 2025 13:39:13 GMT"
},
{
"version": "v2",
"created": "Tue, 29 Apr 2025 11:47:26 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Roychowdhury",
"Dibakar",
""
]
] |
2502.10872 | Qixin Liang | Qixin Liang | Corotational Hinge-based Thin Plates/Shells | Accepted at Eurographics 2025 | Comput. Graph. Forum 44(2), e70022 (Proc. Eurographics 2025) | 10.1111/cgf.70022 | null | cs.GR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present six thin plate/shell models, derived from three distinct types of curvature operators formulated within the corotational frame, for simulating both rest-flat and rest-curved triangular meshes. Each curvature operator derives a curvature expression corresponding to both a plate model and a shell model. The corotational edge-based hinge model uses an edge-based stencil to compute directional curvature, while the corotational FVM hinge model utilizes a triangle-centered stencil, applying the finite volume method (FVM) to superposition directional curvatures across edges, yielding a generalized curvature. The corotational smoothed hinge model also employs a triangle-centered stencil but transforms directional curvatures into a generalized curvature based on a quadratic surface fit. All models assume small strain and small curvature, leading to constant bending energy Hessians, which benefit implicit integrators. Through quantitative benchmarks and qualitative elastodynamic simulations with large time steps, we demonstrate the accuracy, efficiency, and stability of these models. Our contributions enhance the thin plate/shell library for use in both computer graphics and engineering applications. | [
{
"version": "v1",
"created": "Sat, 15 Feb 2025 18:09:10 GMT"
},
{
"version": "v2",
"created": "Tue, 18 Feb 2025 04:54:10 GMT"
},
{
"version": "v3",
"created": "Wed, 05 Mar 2025 08:05:03 GMT"
},
{
"version": "v4",
"created": "Sat, 15 Mar 2025 14:05:40 GMT"
},
{
"version": "v5",
"created": "Mon, 28 Apr 2025 13:30:21 GMT"
},
{
"version": "v6",
"created": "Thu, 05 Jun 2025 02:55:51 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Liang",
"Qixin",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.