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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2501.17438 | Wei Li | Wei Li, Alberto F. Mart\'in, Santiago Badia | Unfitted finite element interpolated neural networks | null | null | null | null | math.NA cs.NA | http://creativecommons.org/licenses/by/4.0/ | We present a novel approach that integrates unfitted finite element methods and neural networks to approximate partial differential equations on complex geometries. Easy-to-generate background meshes (e.g., a simple Cartesian mesh) that cut the domain boundary (i.e., they do not conform to it) are used to build suitable trial and test finite element spaces. The method seeks a neural network that, when interpolated onto the trial space, minimises a discrete norm of the weak residual functional on the test space associated to the equation. As with unfitted finite elements, essential boundary conditions are weakly imposed by Nitsche's method. The method is robust to variations in Nitsche coefficient values, and to small cut cells. We experimentally demonstrate the method's effectiveness in solving both forward and inverse problems across various 2D and 3D complex geometries, including those defined by implicit level-set functions and explicit stereolithography meshes. For forward problems with smooth analytical solutions, the trained neural networks achieve several orders of magnitude smaller $H^1$ errors compared to their interpolation counterparts. These interpolations also maintain expected $h$- and $p$-convergence rates. Using the same amount of training points, the method is faster than standard PINNs (on both GPU and CPU architectures) while achieving similar or superior accuracy. Moreover, using a discrete dual norm of the residual (achieved by cut cell stabilisation) remarkably accelerates neural network training and further enhances robustness to the choice of Nitsche coefficient values. The experiments also show the method's high accuracy and reliability in solving inverse problems, even with incomplete observations. | [
{
"version": "v1",
"created": "Wed, 29 Jan 2025 06:36:01 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 23:45:21 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Li",
"Wei",
""
],
[
"Martín",
"Alberto F.",
""
],
[
"Badia",
"Santiago",
""
]
] |
2501.17509 | Ramon Brasser | R. Brasser, E. W. Wong and S. C. Werner | The current cratering rate on the regular satellites of Jupiter, Saturn, and Uranus | In revision with Astronomy and Astrophysics | A&A 695, A276 (2025) | 10.1051/0004-6361/202453433 | null | astro-ph.EP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We aim to compute the impact rates for objects with a diameter of 1 km onto the regular satellites of Jupiter, Saturn and Uranus using our latest dynamical simulations of the evolution of outer solar system coupled with the best estimates of the current population of objects beyond Neptune and their size-frequency distribution. We use the outcome of the last 3.5~Gyr of evolution of the outer solar system from our database of simulations and combine this with observational constraints of the population beyond Neptune to compute the flux of objects entering the Centaur region, with uncertainties. The initial conditions resemble the current population rather than a near-circular, near-planar disc usually assumed just before the onset of giant planet migration. We obtain a better estimate of the impact probability of a Centaur with the satellites from enacting simulations of planetesimals flying past the satellites on hyperbolic orbits, which agree with literature precedents. We find that our impact rate of objects greater than 1 km in diameter with Jupiter is 0.0012/yr, which is a factor of 3--6 lower than previous estimates of 0.0044/yr from Nesvorny et al. (2023) and 0.0075/yr from Zahnle et al. (2003). On the other hand our impact probabilities with the satellites scaled to the giant planets are consistent with these earlier literature estimates, as is the leakage rate of objects from beyond Neptune into the Centaur region. However, our absolute impact probabilities with the giant planets are lower. We attribute this to our choice of initial conditions. | [
{
"version": "v1",
"created": "Wed, 29 Jan 2025 09:25:09 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Brasser",
"R.",
""
],
[
"Wong",
"E. W.",
""
],
[
"Werner",
"S. C.",
""
]
] |
2501.17737 | Adrian Hill | Adrian Hill and Guillaume Dalle | Sparser, Better, Faster, Stronger: Sparsity Detection for Efficient Automatic Differentiation | 33 pages, 6 figures, 6 tables, 3 listings | null | null | null | cs.LG cs.MS | http://creativecommons.org/licenses/by/4.0/ | From implicit differentiation to probabilistic modeling, Jacobian and Hessian matrices have many potential use cases in Machine Learning (ML), but they are viewed as computationally prohibitive. Fortunately, these matrices often exhibit sparsity, which can be leveraged to speed up the process of Automatic Differentiation (AD). This paper presents advances in sparsity detection, previously the performance bottleneck of Automatic Sparse Differentiation (ASD). Our implementation of sparsity detection is based on operator overloading, able to detect both local and global sparsity patterns, and supports flexible index set representations. It is fully automatic and requires no modification of user code, making it compatible with existing ML codebases. Most importantly, it is highly performant, unlocking Jacobians and Hessians at scales where they were considered too expensive to compute. On real-world problems from scientific ML, graph neural networks and optimization, we show significant speed-ups of up to three orders of magnitude. Notably, using our sparsity detection system, ASD outperforms standard AD for one-off computations, without amortization of either sparsity detection or matrix coloring. | [
{
"version": "v1",
"created": "Wed, 29 Jan 2025 16:21:54 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 14:56:28 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Hill",
"Adrian",
""
],
[
"Dalle",
"Guillaume",
""
]
] |
2501.18326 | Petr Hlin\v{e}n\'y | Petr Hlin\v{e}n\'y and Jan Jedelsk\'y | Transductions of Graph Classes Admitting Product Structure | null | null | null | null | cs.LO math.CO | http://creativecommons.org/licenses/by-sa/4.0/ | In a quest to thoroughly understand the first-order transduction hierarchy of hereditary graph classes, some questions in particular stand out; such as, what properties hold for graph classes that are first-order transductions of planar graphs (and of similar classes)? When addressing this (so-far wide open) question, we turn to the concept of a product structure - being a subgraph of the strong product of a path and a graph of bounded tree-width, introduced by Dujmovic et al. [JACM 2020]. Namely, we prove that any graph class which is a first-order transduction of a class admitting such product structure, up to perturbations also meets a structural description generalizing the concept of a product structure in a dense hereditary way - the latter concept being introduced just recently by Hlineny and Jedelsky under the name of H-clique-width [MFCS 2024]. Using this characterization, we show that the class of the 3D grids, as well as a class of certain modifications of 2D grids, are not first-order transducible from classes admitting a product structure, and in particular not from the class of planar graphs. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 13:13:28 GMT"
},
{
"version": "v2",
"created": "Fri, 21 Mar 2025 10:45:25 GMT"
},
{
"version": "v3",
"created": "Tue, 27 May 2025 08:13:59 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 14:15:00 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Hliněný",
"Petr",
""
],
[
"Jedelský",
"Jan",
""
]
] |
2501.18384 | T. Serkan Kasirga | T. Serkan Kas{\i}rga and Berke K\"oker | An extensive thermal conductivity measurement method based on atomic force microscopy | null | null | 10.1016/j.ijheatmasstransfer.2025.127394 | null | cond-mat.mes-hall | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Heat transport in low-dimensional solids can significantly differ from their bulk counterpart due to various size-related effects. This offers rich heat transport phenomena to emerge. However, finding an appropriate thermometry method for thermal conductivity measurements at the reduced size and dimensionality of the samples is a challenge. Here, we propose and study the feasibility of a nanoscale resolution thermal conductivity measurement method based on bolometric thermometry implemented on an atomic force microscopy (AFM). The local heat exchange between the AFM tip and the sample occurs at a suspended section of the sample, and thermal modeling of the measured electrical resistance change resulting from the bolometric effect provides a unique value for thermal conductivity. As we illustrate via thermal simulations, the proposed method can measure thermal conductivity with thermal disturbance to the sample in as little as 0.2 K at ~20 nm lateral resolution. Our in-depth analysis shows the feasibility and extensive applicability of the proposed AFM-based bolometric thermometry method on low-dimensional materials both in diffusive and ballistic heat transport regimes from cryogenic to above-room temperature. Consequently, the proposed method can lead to a deeper experimental understanding of fundamental questions in nanoscale and low-dimensional heat transport phenomena in many different material classes, as well as Fourier and non-Fourier heat transfer regimes. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 14:33:02 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Kasırga",
"T. Serkan",
""
],
[
"Köker",
"Berke",
""
]
] |
2501.18423 | Tom Dooney | Tom Dooney, Harsh Narola, Stefano Bromuri, R. Lyana Curier, Chris Van Den Broeck, Sarah Caudill, Daniel Stanley Tan | DeepExtractor: Time-domain reconstruction of signals and glitches in gravitational wave data with deep learning | 24 pages, 17 figures, 3 tables | null | null | null | gr-qc astro-ph.IM cs.LG physics.data-an physics.ins-det | http://creativecommons.org/licenses/by/4.0/ | Gravitational wave (GW) detectors, such as LIGO, Virgo, and KAGRA, detect faint signals from distant astrophysical events. However, their high sensitivity also makes them susceptible to background noise, which can obscure these signals. This noise often includes transient artifacts called 'glitches', that can mimic genuine astrophysical signals or mask their true characteristics. In this study, we present DeepExtractor, a deep learning framework that is designed to reconstruct signals and glitches with power exceeding interferometer noise, regardless of their source. We design DeepExtractor to model the inherent noise distribution of GW detectors, following conventional assumptions that the noise is Gaussian and stationary over short time scales. It operates by predicting and subtracting the noise component of the data, retaining only the clean reconstruction of signal or glitch. We focus on applications related to glitches and validate DeepExtractor's effectiveness through three experiments: (1) reconstructing simulated glitches injected into simulated detector noise, (2) comparing its performance with the state-of-the-art BayesWave algorithm, and (3) analyzing real data from the Gravity Spy dataset to demonstrate effective glitch subtraction from LIGO strain data. We further demonstrate its potential by reconstructing three real GW events from LIGO's third observing run, without being trained on GW waveforms. Our proposed model achieves a median mismatch of only 0.9% for simulated glitches, outperforming several deep learning baselines. Additionally, DeepExtractor surpasses BayesWave in glitch recovery, offering a dramatic computational speedup by reconstructing one glitch sample in approximately 0.1 seconds on a CPU, compared to BayesWave's processing time of approximately one hour per glitch. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 15:25:30 GMT"
},
{
"version": "v2",
"created": "Wed, 19 Mar 2025 13:45:31 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 09:34:01 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Dooney",
"Tom",
""
],
[
"Narola",
"Harsh",
""
],
[
"Bromuri",
"Stefano",
""
],
[
"Curier",
"R. Lyana",
""
],
[
"Broeck",
"Chris Van Den",
""
],
[
"Caudill",
"Sarah",
""
],
[
"Tan",
"Daniel Stanley",
""
]
] |
2501.18541 | Ziquan Yang | Haoyang Guo, Ziquan Yang | The Tate conjecture for surfaces of geometric genus one -- embracing singularities | 53 pages, major revision | null | null | null | math.AG | http://creativecommons.org/licenses/by/4.0/ | In this article, we aim to largely complete the program of proving the Tate conjecture for surfaces of geometric genus one, by introducing techniques to analyze those surfaces whose "natural models" are singular. As an application, we show that every elliptic curve of height one over a global function field of genus one and characteristic $p \ge 11$ satisfies the Birch--Swinnerton-Dyer conjecture. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 18:08:44 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:10:42 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Guo",
"Haoyang",
""
],
[
"Yang",
"Ziquan",
""
]
] |
2501.18744 | Andrew Sills | Robert Schneider, Andrew V. Sills, Hunter Waldron | On the $q$-factorization of power series | 11 pages. This is the revision that was accepted for publication in the Ramanujan Journal | null | null | null | math.CO math.NT | http://creativecommons.org/licenses/by/4.0/ | Any power series with unit constant term can be factored into an infinite product of the form $\prod_{n\geq 1} (1-q^n)^{-a_n}$. We give direct formulas for the exponents $a_n$ in terms of the coefficients of the power series, and vice versa, as sums over partitions. As examples, we prove identities for certain partition enumeration functions. Finally, we note $q$-analogues of our enumeration formulas. | [
{
"version": "v1",
"created": "Thu, 30 Jan 2025 20:49:13 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 16:58:45 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Schneider",
"Robert",
""
],
[
"Sills",
"Andrew V.",
""
],
[
"Waldron",
"Hunter",
""
]
] |
2501.19169 | Travis S. Metcalfe | Travis S. Metcalfe, Pascal Petit, Jennifer L. van Saders, Thomas R. Ayres, Derek Buzasi, Oleg Kochukhov, Keivan G. Stassun, Marc H. Pinsonneault, Ilya V. Ilyin, Klaus G. Strassmeier, Adam J. Finley, Rafael A. Garcia, Daniel Huber, Yuxi Lucy Lu, Victor See | Testing the Rossby Paradigm: Weakened Magnetic Braking in early K-type Stars | ApJ accepted, 13 pages including 6 figures and 1 table. Python code is available at https://github.com/travismetcalfe/FinleyMatt2018 | Astrophys. J. 986, 120 (2025) | 10.3847/1538-4357/add40a | null | astro-ph.SR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | There is an intricate relationship between the organization of large-scale magnetic fields by a stellar dynamo and the rate of angular momentum loss due to magnetized stellar winds. An essential ingredient for the operation of a large-scale dynamo is the Coriolis force, which imprints organizing flows on the global convective patterns and inhibits the complete cancellation of bipolar magnetic regions. Consequently, it is natural to expect a rotational threshold for large-scale dynamo action and for the efficient angular momentum loss that it mediates through magnetic braking. Here we present new observational constraints on magnetic braking for an evolutionary sequence of six early K-type stars. To determine the wind braking torque for each of our targets, we combine spectropolarimetric constraints on the large-scale magnetic field, Ly-alpha or X-ray constraints on the mass-loss rate, as well as uniform estimates of the stellar rotation period, mass, and radius. As identified previously from similar observations of hotter stars, we find that the wind braking torque decreases abruptly by more than an order of magnitude at a critical value of the stellar Rossby number. Given that all of the stars in our sample exhibit clear activity cycles, we suggest that weakened magnetic braking may coincide with the operation of a subcritical stellar dynamo. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 14:34:44 GMT"
},
{
"version": "v2",
"created": "Mon, 31 Mar 2025 17:12:53 GMT"
},
{
"version": "v3",
"created": "Mon, 28 Apr 2025 19:45:56 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Metcalfe",
"Travis S.",
""
],
[
"Petit",
"Pascal",
""
],
[
"van Saders",
"Jennifer L.",
""
],
[
"Ayres",
"Thomas R.",
""
],
[
"Buzasi",
"Derek",
""
],
[
"Kochukhov",
"Oleg",
""
],
[
"Stassun",
"Keivan G.",
""
],
[
"Pinsonneault",
"Marc H.",
""
],
[
"Ilyin",
"Ilya V.",
""
],
[
"Strassmeier",
"Klaus G.",
""
],
[
"Finley",
"Adam J.",
""
],
[
"Garcia",
"Rafael A.",
""
],
[
"Huber",
"Daniel",
""
],
[
"Lu",
"Yuxi Lucy",
""
],
[
"See",
"Victor",
""
]
] |
2502.00157 | Rahul Deshpande | Mitchell Lozier, Rahul Deshpande, Ahmad Zarei, Luka Lindi\'c, Wagih Abu Rowin and Ivan Marusic | Defining the mean turbulent boundary layer thickness based on streamwise velocity skewness | Manuscript submitted after first revision to the Journal of Fluid Mechanics, with 26 pages and 10 figures | null | null | null | physics.flu-dyn | http://creativecommons.org/licenses/by-nc-nd/4.0/ | A new statistical definition for the mean turbulent boundary layer thickness is introduced, based on identification of the point where the streamwise velocity skewness changes sign, from negative to positive, in the outermost region of the boundary layer. Importantly, this definition is independent of arbitrary thresholds, and broadly applicable, including to past single-point measurements. Further, this definition is motivated by the phenomenology of streamwise velocity fluctuations near the turbulent/non-turbulent interface, whose local characteristics are shown to be universal for turbulent boundary layers under low freestream turbulence conditions (i.e., with or without pressure gradients, surface roughness, etc.) through large-scale experiments, simulations and coherent structure-based modelling. The new approach yields a turbulent boundary layer thickness that is consistent with previous definitions, such as those based on Reynolds shear stress or `composite' mean velocity profiles, and which can be used practically e.g., to calculate integral thicknesses. Two methods are proposed for estimating the turbulent boundary layer thickness using this definition: one based on simple linear interpolation and the other on fitting a generalised Fourier model to the outer skewness profile. The robustness and limitations of these methods are demonstrated through analysis of several published experimental and numerical datasets, which cover a range of canonical and non-canonical turbulent boundary layers. These datasets also vary in key characteristics such as wall-normal resolution and measurement noise, particularly in the critical turbulent/non-turbulent interface region. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 20:47:14 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 09:01:31 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Lozier",
"Mitchell",
""
],
[
"Deshpande",
"Rahul",
""
],
[
"Zarei",
"Ahmad",
""
],
[
"Lindić",
"Luka",
""
],
[
"Rowin",
"Wagih Abu",
""
],
[
"Marusic",
"Ivan",
""
]
] |
2502.00373 | Amy Xiang Wang | Amy Xiang Wang, Zakhar Shumaylov, Peter Zaika, Ferdia Sherry, Carola-Bibiane Sch\"onlieb | Generalized Lie Symmetries in Physics-Informed Neural Operators | COLT 2025 Theory of AI for Scientific Computing Workshop Best Paper Runner-Up Award; SCML 2025 Oral | null | null | null | cs.LG physics.comp-ph | http://creativecommons.org/licenses/by/4.0/ | Physics-informed neural operators (PINOs) have emerged as powerful tools for learning solution operators of partial differential equations (PDEs). Recent research has demonstrated that incorporating Lie point symmetry information can significantly enhance the training efficiency of PINOs, primarily through techniques like data, architecture, and loss augmentation. In this work, we focus on the latter, highlighting that point symmetries oftentimes result in no training signal, limiting their effectiveness in many problems. To address this, we propose a novel loss augmentation strategy that leverages evolutionary representatives of point symmetries, a specific class of generalized symmetries of the underlying PDE. These generalized symmetries provide a richer set of generators compared to standard symmetries, leading to a more informative training signal. We demonstrate that leveraging evolutionary representatives enhances the performance of neural operators, resulting in improved data efficiency and accuracy during training. | [
{
"version": "v1",
"created": "Sat, 01 Feb 2025 09:21:07 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 03:08:57 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Wang",
"Amy Xiang",
""
],
[
"Shumaylov",
"Zakhar",
""
],
[
"Zaika",
"Peter",
""
],
[
"Sherry",
"Ferdia",
""
],
[
"Schönlieb",
"Carola-Bibiane",
""
]
] |
2502.00479 | Ahmed Zubair | Md. Ehsanul Karim and Ahmed Zubair | Star-shaped nanostructure-based multifunctional chiral metasurface for near-infrared wavelengths | null | null | 10.1016/j.optlaseng.2025.109149 | null | physics.optics physics.app-ph physics.comp-ph | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Here, we reported, for the first time, a chiral metasurface with multifunctional capabilities tailored for the near-infrared (NIR) wavelength range, garnering tremendous interest in applications such as optical communication and chiral sensing. The structure comprised star and bar-shaped silver antennas immersed in SiO2. In computational simulations based on the finite-difference time-domain method, the proposed architecture yielded a maximum circular dichroism of 0.7 at 971.5 nm wavelength and a maximum asymmetric transmission under linearly polarized excitation of 0.92 at 1020 nm wavelength. Moreover, this multifunctional structure manifested near-perfect cross-polarization conversions for both linear and circular polarizations, as well as linear to circular and elliptical polarization conversion effects at different bands within the NIR spectral region. The enhanced light-matter interaction of the proposed structure was utilized to detect and differentiate two enantiomers of 1,2-propanediol, yielding excellent chirality sensing capability. As a multifunctional compact chiral platform, the proposed structure is expected to find direct applications in integrated and miniaturized multi-channel optical communication platforms, materials and biomolecules sensing in attomolar concentrations, and the pharmaceutical industry. | [
{
"version": "v1",
"created": "Sat, 01 Feb 2025 16:06:53 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Karim",
"Md. Ehsanul",
""
],
[
"Zubair",
"Ahmed",
""
]
] |
2502.00963 | Mauricio Soroco | Mauricio Soroco, Jialin Song, Mengzhou Xia, Kye Emond, Weiran Sun, Wuyang Chen | PDE-Controller: LLMs for Autoformalization and Reasoning of PDEs | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | While recent AI-for-math has made strides in pure mathematics, areas of applied mathematics, particularly PDEs, remain underexplored despite their significant real-world applications. We present PDE-Controller, a framework that enables large language models (LLMs) to control systems governed by partial differential equations (PDEs). Our approach enables LLMs to transform informal natural language instructions into formal specifications, and then execute reasoning and planning steps to improve the utility of PDE control. We build a holistic solution comprising datasets (both human-written cases and 2 million synthetic samples), math-reasoning models, and novel evaluation metrics, all of which require significant effort. Our PDE-Controller significantly outperforms prompting the latest open source and GPT models in reasoning, autoformalization, and program synthesis, achieving up to a 62% improvement in utility gain for PDE control. By bridging the gap between language generation and PDE systems, we demonstrate the potential of LLMs in addressing complex scientific and engineering challenges. We release all data, model checkpoints, and code at https://pde-controller.github.io/. | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 00:03:41 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 03:07:41 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Soroco",
"Mauricio",
""
],
[
"Song",
"Jialin",
""
],
[
"Xia",
"Mengzhou",
""
],
[
"Emond",
"Kye",
""
],
[
"Sun",
"Weiran",
""
],
[
"Chen",
"Wuyang",
""
]
] |
2502.01340 | Tim Donkers | Tim Donkers, J\"urgen Ziegler | Human-Agent Interaction in Synthetic Social Networks: A Framework for Studying Online Polarization | null | null | null | null | physics.soc-ph cs.SI | http://creativecommons.org/licenses/by/4.0/ | Online social networks have dramatically altered the landscape of public discourse, creating both opportunities for enhanced civic participation and risks of deepening social divisions. Prevalent approaches to studying online polarization have been limited by a methodological disconnect: mathematical models excel at formal analysis but lack linguistic realism, while language model-based simulations capture natural discourse but often sacrifice analytical precision. This paper introduces an innovative computational framework that synthesizes these approaches by embedding formal opinion dynamics principles within LLM-based artificial agents, enabling both rigorous mathematical analysis and naturalistic social interactions. We validate our framework through comprehensive offline testing and experimental evaluation with 122 human participants engaging in a controlled social network environment. The results demonstrate our ability to systematically investigate polarization mechanisms while preserving ecological validity. Our findings reveal how polarized environments shape user perceptions and behavior: participants exposed to polarized discussions showed markedly increased sensitivity to emotional content and group affiliations, while perceiving reduced uncertainty in the agents' positions. By combining mathematical precision with natural language capabilities, our framework opens new avenues for investigating social media phenomena through controlled experimentation. This methodological advancement allows researchers to bridge the gap between theoretical models and empirical observations, offering unprecedented opportunities to study the causal mechanisms underlying online opinion dynamics. | [
{
"version": "v1",
"created": "Mon, 03 Feb 2025 13:34:00 GMT"
},
{
"version": "v2",
"created": "Wed, 05 Mar 2025 14:35:51 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 10:27:42 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Donkers",
"Tim",
""
],
[
"Ziegler",
"Jürgen",
""
]
] |
2502.01916 | Tim-Lukas Habich | Tim-Lukas Habich, Aran Mohammad, Simon F. G. Ehlers, Martin Bensch, Thomas Seel, Moritz Schappler | Generalizable and Fast Surrogates: Model Predictive Control of Articulated Soft Robots using Physics-Informed Neural Networks | null | null | null | null | cs.RO cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Soft robots can revolutionize several applications with high demands on dexterity and safety. When operating these systems, real-time estimation and control require fast and accurate models. However, prediction with first-principles (FP) models is slow, and learned black-box models have poor generalizability. Physics-informed machine learning offers excellent advantages here, but it is currently limited to simple, often simulated systems without considering changes after training. We propose physics-informed neural networks (PINNs) for articulated soft robots (ASRs) with a focus on data efficiency. The amount of expensive real-world training data is reduced to a minimum -- one dataset in one system domain. Two hours of data in different domains are used for a comparison against two gold-standard approaches: In contrast to a recurrent neural network, the PINN provides a high generalizability. The prediction speed of an accurate FP model is exceeded with the PINN by up to a factor of 467 at slightly reduced accuracy. This enables nonlinear model predictive control (MPC) of a pneumatic ASR. Accurate position tracking with the MPC running at 47 Hz is achieved in six dynamic experiments. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 01:16:33 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 12:07:14 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Habich",
"Tim-Lukas",
""
],
[
"Mohammad",
"Aran",
""
],
[
"Ehlers",
"Simon F. G.",
""
],
[
"Bensch",
"Martin",
""
],
[
"Seel",
"Thomas",
""
],
[
"Schappler",
"Moritz",
""
]
] |
2502.01920 | Yalin Liao | Yalin Liao and Austin J. Brockmeier | Anomaly Detection via Autoencoder Composite Features and NCE | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Unsupervised anomaly detection is a challenging task. Autoencoders (AEs) or generative models are often employed to model the data distribution of normal inputs and subsequently identify anomalous, out-of-distribution inputs by high reconstruction error or low likelihood, respectively. However, AEs may generalize and achieve small reconstruction errors on abnormal inputs. We propose a decoupled training approach for anomaly detection that both an AE and a likelihood model trained with noise contrastive estimation (NCE). After training the AE, NCE estimates a probability density function, to serve as the anomaly score, on the joint space of the AE's latent representation combined with features of the reconstruction quality. To further reduce the false negative rate in NCE we systematically varying the reconstruction features to augment the training and optimize the contrastive Gaussian noise distribution. Experimental assessments on multiple benchmark datasets demonstrate that the proposed approach matches the performance of prevalent state-of-the-art anomaly detection algorithms. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 01:29:22 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 14:05:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Liao",
"Yalin",
""
],
[
"Brockmeier",
"Austin J.",
""
]
] |
2502.02105 | S. V. Zaitsev-Zotov | R.M. Lukmanova, I.A. Cohn, V.E. Minakova, and S.V. Zaitsev-Zotov | Strain-induced proximity effect in topological insulator TaSe$_3$ | 5 pages, 6 figures, to be published in PRB (2025) | null | null | null | cond-mat.supr-con cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The magnetoresistance of superconductor-topological insulator-superconductor structures, with indium as the superconductor and TaSe$_3$ as the topological insulator, shows steplike features on the resistance under magnetic fields. These resistance steps are resulted from the suppression of superconductivity, induced by the superconducting proximity effect in both the bulk and surface states of the topological insulator. The position and amplitude of the steps, occurring at approximately 0.1 T, show an unusual dependence on the magnitude of the uniaxial strain ($\epsilon$), indicating their connection with surface states. This behavior follows the expected transition sequence: semi-metal $\rightarrow$ strong topological insulator $\rightarrow$ trivial insulator, and supports the presence of surface states at $0.46\% \lesssim \epsilon \lesssim 0.85\%$. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 08:38:05 GMT"
},
{
"version": "v2",
"created": "Tue, 22 Apr 2025 14:09:14 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 08:43:09 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Lukmanova",
"R. M.",
""
],
[
"Cohn",
"I. A.",
""
],
[
"Minakova",
"V. E.",
""
],
[
"Zaitsev-Zotov",
"S. V.",
""
]
] |
2502.02157 | Olga Sapir | Olga B. Sapir | A sufficient condition under which a monoid is non-finitely related, and a sufficient condition under which a monoid is finitely related | Added a sufficient condition under which a monoid is finitely related and used it to show that M(aabb) is finitely related | null | null | null | math.GR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We transform the method of Glasson into a sufficient condition under which a monoid is non-finitely related, add a new member to the collection of interlocking word-patterns, and use it to show that the monoid $M(ab^2a, a^2b^2)$ is non-finitely related.
We also give a sufficient condition under which a monoid is finitely related and use it show that $M(a^2b^2)$ is finitely related. Together with the results of Glasson this completes the description of all finitely related monoids among the monoids of the form $M(W)$ where every word ${\bf u} \in W$ depends on two variables and every variable occurs twice in $\bf u$. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 09:35:25 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 07:00:59 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Sapir",
"Olga B.",
""
]
] |
2502.02221 | Ji\v{r}\'i N\v{e}me\v{c}ek | Ji\v{r}\'i N\v{e}me\v{c}ek, Mark Kozdoba, Illia Kryvoviaz, Tom\'a\v{s} Pevn\'y, Jakub Mare\v{c}ek | Bias Detection via Maximum Subgroup Discrepancy | 12 pages, 6 figures | Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (2025) | 10.1145/3711896.3736857 | null | cs.LG cs.AI stat.ML | http://creativecommons.org/licenses/by/4.0/ | Bias evaluation is fundamental to trustworthy AI, both in terms of checking data quality and in terms of checking the outputs of AI systems. In testing data quality, for example, one may study the distance of a given dataset, viewed as a distribution, to a given ground-truth reference dataset. However, classical metrics, such as the Total Variation and the Wasserstein distances, are known to have high sample complexities and, therefore, may fail to provide a meaningful distinction in many practical scenarios.
In this paper, we propose a new notion of distance, the Maximum Subgroup Discrepancy (MSD). In this metric, two distributions are close if, roughly, discrepancies are low for all feature subgroups. While the number of subgroups may be exponential, we show that the sample complexity is linear in the number of features, thus making it feasible for practical applications. Moreover, we provide a practical algorithm for evaluating the distance based on Mixed-integer optimization (MIO). We also note that the proposed distance is easily interpretable, thus providing clearer paths to fixing the biases once they have been identified. Finally, we describe a natural general bias detection framework, termed MSDD distances, and show that MSD aligns well with this framework. We empirically evaluate MSD by comparing it with other metrics and by demonstrating the above properties of MSD on real-world datasets. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 11:01:03 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 08:50:00 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Němeček",
"Jiří",
""
],
[
"Kozdoba",
"Mark",
""
],
[
"Kryvoviaz",
"Illia",
""
],
[
"Pevný",
"Tomáš",
""
],
[
"Mareček",
"Jakub",
""
]
] |
2502.02697 | Carsten M\"uller | S. Remme, A. B. Voitkiv, G. Pretzler and C. M\"uller | Phenomenological rate formulas for over-barrier ionization of hydrogen and helium atoms in strong constant electric fields | 9 pages, 6 figures | null | null | null | physics.atom-ph physics.plasm-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Nonrelativistic over-barrier ionization (OBI) of atoms in strong electric fields is studied, focussing on hydrogen and helium as concrete examples. Our goal is, on the one hand, to develop an intuitive physical picture behind established empirical formulas for the ionization rate. We show that the ionization rate in a near OBI regime can be modelled quantitatively by extending corresponding tunneling rates by the combined action of the Stark effect and a widened electron emission angle. On the other hand, we present analytical rate formulas in a far OBI regime which closely agree with available numerical data. In result, compact rate expressions describing OBI of hydrogen-like and helium atoms in a broad range of applied field strengths are obtained. They can be useful, for example, in numerical laser-plasma simulation codes to describe elementary ionization events. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 20:26:10 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 11:19:12 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Remme",
"S.",
""
],
[
"Voitkiv",
"A. B.",
""
],
[
"Pretzler",
"G.",
""
],
[
"Müller",
"C.",
""
]
] |
2502.02747 | Yuheng Tang | Hongwei Li, Yuheng Tang, Shiqi Wang, Wenbo Guo | PatchPilot: A Cost-Efficient Software Engineering Agent with Early Attempts on Formal Verification | null | null | null | null | cs.RO cs.AI cs.CR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent research builds various patching agents that combine large language models (LLMs) with non-ML tools and achieve promising results on the state-of-the-art (SOTA) software patching benchmark, SWE-bench. Based on how to determine the patching workflows, existing patching agents can be categorized as agent-based planning methods, which rely on LLMs for planning, and rule-based planning methods, which follow a pre-defined workflow. At a high level, agent-based planning methods achieve high patching performance but with a high cost and limited stability. Rule-based planning methods, on the other hand, are more stable and efficient but have key workflow limitations that compromise their patching performance. In this paper, we propose PatchPilot, an agentic patcher that strikes a balance between patching efficacy, stability, and cost-efficiency. PatchPilot proposes a novel rule-based planning workflow with five components: reproduction, localization, generation, validation, and refinement (where refinement is unique to PatchPilot). We introduce novel and customized designs to each component to optimize their effectiveness and efficiency. Through extensive experiments on the SWE-bench benchmarks, PatchPilot shows a superior performance than existing open-source methods while maintaining low cost (less than 1$ per instance) and ensuring higher stability. We also conduct a detailed ablation study to validate the key designs in each component. Our code is available at https://github.com/ucsb-mlsec/PatchPilot. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 22:30:02 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 18:19:40 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Li",
"Hongwei",
""
],
[
"Tang",
"Yuheng",
""
],
[
"Wang",
"Shiqi",
""
],
[
"Guo",
"Wenbo",
""
]
] |
2502.04138 | Aravind Plathanam Babu | Aravind Plathanam Babu, Oskari Kerppo, Andr\'es Mu\~noz Moller, Majid Haghparast, Matti Silveri | Gate teleportation-assisted routing for quantum algorithms | null | null | 10.1088/2058-9565/adcae4 | null | quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The limited qubit connectivity of quantum processors poses a significant challenge in deploying practical algorithms and logical gates, necessitating efficient qubit mapping and routing strategies. When implementing a gate that requires additional connectivity beyond the native connectivity, the qubit state must be moved to a nearby connected qubit to execute the desired gate locally. This is typically achieved using a series of SWAP gates creating a SWAP path. However, routing methods relying on SWAP gates often lead to increased circuit depth and gate count, motivating the need for alternative approaches. This work explores the potential of teleported gates to improve qubit routing efficiency, focusing on implementation within specific hardware topologies and benchmark quantum algorithms. We propose a routing method that is assisted by gate teleportation. It establishes additional connectivity using gate teleportation paths through available unused qubits, termed auxiliary qubits, within the topology. To optimize this approach, we have developed an algorithm to identify the best gate teleportation connections, considering their potential to reduce the depth of the circuit and address possible errors that may arise from the teleportation paths. Finally, we demonstrate depth reduction with gate teleportation-assisted routing in various benchmark algorithms, including case studies on the compilation of the Deutsch-Jozsa algorithm and the Quantum Approximation Optimization Algorithm (QAOA) for heavy-hexagon topology used in IBM 127-qubit Eagle r3 processors. Our benchmark results show a 10-25 $\%$ depth reduction in the routing of selected algorithms compared to regular routing without using the teleported gate. | [
{
"version": "v1",
"created": "Thu, 06 Feb 2025 15:18:13 GMT"
},
{
"version": "v2",
"created": "Mon, 14 Apr 2025 15:24:12 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Babu",
"Aravind Plathanam",
""
],
[
"Kerppo",
"Oskari",
""
],
[
"Moller",
"Andrés Muñoz",
""
],
[
"Haghparast",
"Majid",
""
],
[
"Silveri",
"Matti",
""
]
] |
2502.04388 | Hepeng Li | Hepeng Li, Yuhong Liu, Jun Yan, Jie Gao, Xiaoou Yang, Mohamed Naili | Position: Emergent Machina Sapiens Urge Rethinking Multi-Agent Paradigms | null | null | null | null | cs.MA cs.AI | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Artificial Intelligence (AI) agents capable of autonomous learning and independent decision-making hold great promise for addressing complex challenges across various critical infrastructure domains, including transportation, energy systems, and manufacturing. However, the surge in the design and deployment of AI systems, driven by various stakeholders with distinct and unaligned objectives, introduces a crucial challenge: How can uncoordinated AI systems coexist and evolve harmoniously in shared environments without creating chaos or compromising safety? To address this, we advocate for a fundamental rethinking of existing multi-agent frameworks, such as multi-agent systems and game theory, which are largely limited to predefined rules and static objective structures. We posit that AI agents should be empowered to adjust their objectives dynamically, make compromises, form coalitions, and safely compete or cooperate through evolving relationships and social feedback. Through two case studies in critical infrastructure applications, we call for a shift toward the emergent, self-organizing, and context-aware nature of these multi-agentic AI systems. | [
{
"version": "v1",
"created": "Wed, 05 Feb 2025 22:20:15 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 23:51:59 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Li",
"Hepeng",
""
],
[
"Liu",
"Yuhong",
""
],
[
"Yan",
"Jun",
""
],
[
"Gao",
"Jie",
""
],
[
"Yang",
"Xiaoou",
""
],
[
"Naili",
"Mohamed",
""
]
] |
2502.04495 | Shurui Gui | Shurui Gui, Xiner Li, Shuiwang Ji | Discovering Physics Laws of Dynamical Systems via Invariant Function Learning | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We consider learning underlying laws of dynamical systems governed by ordinary differential equations (ODE). A key challenge is how to discover intrinsic dynamics across multiple environments while circumventing environment-specific mechanisms. Unlike prior work, we tackle more complex environments where changes extend beyond function coefficients to entirely different function forms. For example, we demonstrate the discovery of ideal pendulum's natural motion $\alpha^2 \sin{\theta_t}$ by observing pendulum dynamics in different environments, such as the damped environment $\alpha^2 \sin(\theta_t) - \rho \omega_t$ and powered environment $\alpha^2 \sin(\theta_t) + \rho \frac{\omega_t}{\left|\omega_t\right|}$. Here, we formulate this problem as an \emph{invariant function learning} task and propose a new method, known as \textbf{D}isentanglement of \textbf{I}nvariant \textbf{F}unctions (DIF), that is grounded in causal analysis. We propose a causal graph and design an encoder-decoder hypernetwork that explicitly disentangles invariant functions from environment-specific dynamics. The discovery of invariant functions is guaranteed by our information-based principle that enforces the independence between extracted invariant functions and environments. Quantitative comparisons with meta-learning and invariant learning baselines on three ODE systems demonstrate the effectiveness and efficiency of our method. Furthermore, symbolic regression explanation results highlight the ability of our framework to uncover intrinsic laws. Our code has been released as part of the AIRS library (\href{https://github.com/divelab/AIRS/tree/main/OpenODE/DIF}{https://github.com/divelab/AIRS/}). | [
{
"version": "v1",
"created": "Thu, 06 Feb 2025 20:46:50 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 06:44:03 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Gui",
"Shurui",
""
],
[
"Li",
"Xiner",
""
],
[
"Ji",
"Shuiwang",
""
]
] |
2502.04959 | Daniel Marczak | Daniel Marczak, Simone Magistri, Sebastian Cygert, Bart{\l}omiej Twardowski, Andrew D. Bagdanov, Joost van de Weijer | No Task Left Behind: Isotropic Model Merging with Common and Task-Specific Subspaces | Accepted at ICML 2025 | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Model merging integrates the weights of multiple task-specific models into a single multi-task model. Despite recent interest in the problem, a significant performance gap between the combined and single-task models remains. In this paper, we investigate the key characteristics of task matrices -- weight update matrices applied to a pre-trained model -- that enable effective merging. We show that alignment between singular components of task-specific and merged matrices strongly correlates with performance improvement over the pre-trained model. Based on this, we propose an isotropic merging framework that flattens the singular value spectrum of task matrices, enhances alignment, and reduces the performance gap. Additionally, we incorporate both common and task-specific subspaces to further improve alignment and performance. Our proposed approach achieves state-of-the-art performance on vision and language tasks across various sets of tasks and model scales. This work advances the understanding of model merging dynamics, offering an effective methodology to merge models without requiring additional training. Code is available at https://github.com/danielm1405/iso-merging . | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 14:22:56 GMT"
},
{
"version": "v2",
"created": "Fri, 30 May 2025 15:26:33 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 12:14:36 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Marczak",
"Daniel",
""
],
[
"Magistri",
"Simone",
""
],
[
"Cygert",
"Sebastian",
""
],
[
"Twardowski",
"Bartłomiej",
""
],
[
"Bagdanov",
"Andrew D.",
""
],
[
"van de Weijer",
"Joost",
""
]
] |
2502.05003 | Andrei Panferov | Andrei Panferov, Jiale Chen, Soroush Tabesh, Roberto L. Castro, Mahdi Nikdan and Dan Alistarh | QuEST: Stable Training of LLMs with 1-Bit Weights and Activations | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | One approach to reducing the massive costs of large language models (LLMs) is the use of quantized or sparse representations for training or deployment. While post-training compression methods are very popular, the question of obtaining even more accurate compressed models by directly training over such representations, i.e., Quantization-Aware Training (QAT), is still open: for example, a recent study (arXiv:2411.04330) put the "optimal" bit-width at which models can be trained using QAT, while staying accuracy-competitive with standard FP16/BF16 precision, at 8-bits weights and activations. We advance this state-of-the-art via a new method called QuEST, for which we demonstrate optimality at 4-bits and stable convergence as low as 1-bit weights and activations. QuEST achieves this by improving two key aspects of QAT methods: (1) accurate and fast quantization of the (continuous) distributions of weights and activations via Hadamard normalization and MSE-optimal fitting; (2) a new trust gradient estimator based on the idea of explicitly minimizing the error between the noisy gradient computed over quantized states and the "true" (but unknown) full-precision gradient. Experiments on Llama-type architectures show that QuEST induces stable scaling laws across the entire range of hardware-supported precisions, and can be extended to sparse representations. We provide GPU kernel support showing that models produced by QuEST can be executed efficiently. Our code is available at https://github.com/IST-DASLab/QuEST. | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 15:23:34 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 18:01:40 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Panferov",
"Andrei",
""
],
[
"Chen",
"Jiale",
""
],
[
"Tabesh",
"Soroush",
""
],
[
"Castro",
"Roberto L.",
""
],
[
"Nikdan",
"Mahdi",
""
],
[
"Alistarh",
"Dan",
""
]
] |
2502.05075 | Yijun Dong | Yijun Dong, Yicheng Li, Yunai Li, Jason D. Lee, Qi Lei | Discrepancies are Virtue: Weak-to-Strong Generalization through Lens of Intrinsic Dimension | ICML 2025 | null | null | null | cs.LG cs.NA math.NA stat.ML | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Weak-to-strong (W2S) generalization is a type of finetuning (FT) where a strong (large) student model is trained on pseudo-labels generated by a weak teacher. Surprisingly, W2S FT often outperforms the weak teacher. We seek to understand this phenomenon through the observation that FT often occurs in intrinsically low-dimensional spaces. Leveraging the low intrinsic dimensionality of FT, we analyze W2S in the ridgeless regression setting from a variance reduction perspective. For a strong student-weak teacher pair with sufficiently expressive low-dimensional feature subspaces $\mathcal{V}_s, \mathcal{V}_w$, we provide an exact characterization of the variance that dominates the generalization error of W2S. This unveils a virtue of discrepancy between the strong and weak models in W2S: the variance of the weak teacher is inherited by the strong student in $\mathcal{V}_s \cap \mathcal{V}_w$, while reduced by a factor of $\mathrm{dim}(\mathcal{V}_s)/N$ in the subspace of discrepancy $\mathcal{V}_w \setminus \mathcal{V}_s$ with $N$ pseudo-labels for W2S. Our analysis further casts light on the sample complexities and the scaling of performance gap recovery in W2S. The analysis is supported by experiments on synthetic regression problems, as well as real vision and NLP tasks. | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 16:46:43 GMT"
},
{
"version": "v2",
"created": "Fri, 11 Apr 2025 19:22:00 GMT"
},
{
"version": "v3",
"created": "Thu, 22 May 2025 15:55:30 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 12:15:49 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Dong",
"Yijun",
""
],
[
"Li",
"Yicheng",
""
],
[
"Li",
"Yunai",
""
],
[
"Lee",
"Jason D.",
""
],
[
"Lei",
"Qi",
""
]
] |
2502.05174 | Kaijie Zhu | Kaijie Zhu, Xianjun Yang, Jindong Wang, Wenbo Guo, William Yang Wang | MELON: Provable Defense Against Indirect Prompt Injection Attacks in AI Agents | ICML 2025 | null | null | null | cs.CR cs.AI | http://creativecommons.org/licenses/by/4.0/ | Recent research has explored that LLM agents are vulnerable to indirect prompt injection (IPI) attacks, where malicious tasks embedded in tool-retrieved information can redirect the agent to take unauthorized actions. Existing defenses against IPI have significant limitations: either require essential model training resources, lack effectiveness against sophisticated attacks, or harm the normal utilities. We present MELON (Masked re-Execution and TooL comparisON), a novel IPI defense. Our approach builds on the observation that under a successful attack, the agent's next action becomes less dependent on user tasks and more on malicious tasks. Following this, we design MELON to detect attacks by re-executing the agent's trajectory with a masked user prompt modified through a masking function. We identify an attack if the actions generated in the original and masked executions are similar. We also include three key designs to reduce the potential false positives and false negatives. Extensive evaluation on the IPI benchmark AgentDojo demonstrates that MELON outperforms SOTA defenses in both attack prevention and utility preservation. Moreover, we show that combining MELON with a SOTA prompt augmentation defense (denoted as MELON-Aug) further improves its performance. We also conduct a detailed ablation study to validate our key designs. Code is available at https://github.com/kaijiezhu11/MELON. | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 18:57:49 GMT"
},
{
"version": "v2",
"created": "Thu, 01 May 2025 20:51:22 GMT"
},
{
"version": "v3",
"created": "Sat, 24 May 2025 23:01:12 GMT"
},
{
"version": "v4",
"created": "Tue, 10 Jun 2025 18:13:09 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Zhu",
"Kaijie",
""
],
[
"Yang",
"Xianjun",
""
],
[
"Wang",
"Jindong",
""
],
[
"Guo",
"Wenbo",
""
],
[
"Wang",
"William Yang",
""
]
] |
2502.05202 | Nadav Timor | Nadav Timor, Jonathan Mamou, Daniel Korat, Moshe Berchansky, Gaurav Jain, Oren Pereg, Moshe Wasserblat, David Harel | Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies | ICML'25 Oral (top %1) | null | null | null | cs.CL cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Accelerating the inference of large language models (LLMs) is a critical challenge in generative AI. Speculative decoding (SD) methods offer substantial efficiency gains by generating multiple tokens using a single target forward pass. However, existing SD approaches require the drafter and target models to share the same vocabulary, thus limiting the pool of possible drafters, often necessitating the training of a drafter from scratch. We present three new SD methods that remove this shared-vocabulary constraint. All three methods preserve the target distribution (i.e., they are lossless) and work with off-the-shelf models without requiring additional training or modifications. Empirically, on summarization, programming, and long-context tasks, our algorithms demonstrate significant speedups of up to 2.8x over standard autoregressive decoding. By enabling any off-the-shelf model to serve as a drafter and requiring no retraining, this work substantially broadens the applicability of the SD framework in practice. | [
{
"version": "v1",
"created": "Fri, 31 Jan 2025 19:13:58 GMT"
},
{
"version": "v2",
"created": "Mon, 09 Jun 2025 20:12:43 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 01:26:45 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Timor",
"Nadav",
""
],
[
"Mamou",
"Jonathan",
""
],
[
"Korat",
"Daniel",
""
],
[
"Berchansky",
"Moshe",
""
],
[
"Jain",
"Gaurav",
""
],
[
"Pereg",
"Oren",
""
],
[
"Wasserblat",
"Moshe",
""
],
[
"Harel",
"David",
""
]
] |
2502.05335 | Roussel Desmond Nzoyem | Roussel Desmond Nzoyem, Grant Stevens, Amarpal Sahota, David A.W. Barton, Tom Deakin | Towards Foundational Models for Dynamical System Reconstruction: Hierarchical Meta-Learning via Mixture of Experts | 22 pages, 11 figures, 7 tables. Accepted as a SCOPE workshop paper at ICLR 2025 | https://openreview.net/forum?id=NteuHm0UXw (2025) | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | As foundational models reshape scientific discovery, a bottleneck persists in dynamical system reconstruction (DSR): the ability to learn across system hierarchies. Many meta-learning approaches have been applied successfully to single systems, but falter when confronted with sparse, loosely related datasets requiring multiple hierarchies to be learned. Mixture of Experts (MoE) offers a natural paradigm to address these challenges. Despite their potential, we demonstrate that naive MoEs are inadequate for the nuanced demands of hierarchical DSR, largely due to their gradient descent-based gating update mechanism which leads to slow updates and conflicted routing during training. To overcome this limitation, we introduce MixER: Mixture of Expert Reconstructors, a novel sparse top-1 MoE layer employing a custom gating update algorithm based on $K$-means and least squares. Extensive experiments validate MixER's capabilities, demonstrating efficient training and scalability to systems of up to ten parametric ordinary differential equations. However, our layer underperforms state-of-the-art meta-learners in high-data regimes, particularly when each expert is constrained to process only a fraction of a dataset composed of highly related data points. Further analysis with synthetic and neuroscientific time series suggests that the quality of the contextual representations generated by MixER is closely linked to the presence of hierarchical structure in the data. | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 21:16:43 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:38:50 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Nzoyem",
"Roussel Desmond",
""
],
[
"Stevens",
"Grant",
""
],
[
"Sahota",
"Amarpal",
""
],
[
"Barton",
"David A. W.",
""
],
[
"Deakin",
"Tom",
""
]
] |
2502.05357 | Kisun Lee | Michael Burr, Michael Byrd, Kisun Lee | Certified algebraic curve projections by path tracking | 23 pages, 7 figures. Accepted for the Proceedings of ISSAC 2025 | null | null | null | cs.SC cs.NA math.AG math.NA | http://creativecommons.org/licenses/by/4.0/ | We present a certified algorithm that takes a smooth algebraic curve in $\mathbb{R}^n$ and computes an isotopic approximation for a generic projection of the curve into $\mathbb{R}^2$. Our algorithm is designed for curves given implicitly by the zeros of $n-1$ polynomials, but it can be partially extended to parametrically defined curves. The main challenge in correctly computing the projection is to guarantee the topological correctness of crossings in the projection. Our approach combines certified path tracking and interval arithmetic in a two-step procedure: first, we construct an approximation to the curve in $\mathbb{R}^n$, and, second, we refine the approximation until the topological correctness of the projection can be guaranteed. We provide a proof-of-concept implementation illustrating the algorithm. | [
{
"version": "v1",
"created": "Fri, 07 Feb 2025 22:08:27 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:17:52 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Burr",
"Michael",
""
],
[
"Byrd",
"Michael",
""
],
[
"Lee",
"Kisun",
""
]
] |
2502.06034 | Mozes Jacobs | Mozes Jacobs, Roberto C. Budzinski, Lyle Muller, Demba Ba, T. Anderson Keller | Traveling Waves Integrate Spatial Information Through Time | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Traveling waves of neural activity are widely observed in the brain, but their precise computational function remains unclear. One prominent hypothesis is that they enable the transfer and integration of spatial information across neural populations. However, few computational models have explored how traveling waves might be harnessed to perform such integrative processing. Drawing inspiration from the famous "Can one hear the shape of a drum?" problem -- which highlights how normal modes of wave dynamics encode geometric information -- we investigate whether similar principles can be leveraged in artificial neural networks. Specifically, we introduce convolutional recurrent neural networks that learn to produce traveling waves in their hidden states in response to visual stimuli, enabling spatial integration. By then treating these wave-like activation sequences as visual representations themselves, we obtain a powerful representational space that outperforms local feed-forward networks on tasks requiring global spatial context. In particular, we observe that traveling waves effectively expand the receptive field of locally connected neurons, supporting long-range encoding and communication of information. We demonstrate that models equipped with this mechanism solve visual semantic segmentation tasks demanding global integration, significantly outperforming local feed-forward models and rivaling non-local U-Net models with fewer parameters. As a first step toward traveling-wave-based communication and visual representation in artificial networks, our findings suggest wave-dynamics may provide efficiency and training stability benefits, while simultaneously offering a new framework for connecting models to biological recordings of neural activity. | [
{
"version": "v1",
"created": "Sun, 09 Feb 2025 21:14:27 GMT"
},
{
"version": "v2",
"created": "Wed, 12 Feb 2025 19:36:57 GMT"
},
{
"version": "v3",
"created": "Mon, 24 Feb 2025 16:47:01 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 15:57:01 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jacobs",
"Mozes",
""
],
[
"Budzinski",
"Roberto C.",
""
],
[
"Muller",
"Lyle",
""
],
[
"Ba",
"Demba",
""
],
[
"Keller",
"T. Anderson",
""
]
] |
2502.06356 | Marco Fuhrman | Marco Fuhrman | The randomization method in stochastic optimal control | arXiv admin note: text overlap with arXiv:1511.09274 | null | null | null | math.OC math.PR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper we make a survey on the so called randomization method, a recent methodology to study stochastic optimization problems. It allows to represent the value function of an optimal control problem by a suitable backward stochastic differential equation (BSDE), by means of an auxiliary optimization problem having the same value as the starting one. This method works for a large class of control problems and provides a BSDE representation to many related PDEs of Hamilton-Jacobi-Bellman type, even in the fully non linear case. After a general informal introduction we explain the method giving full details in a basic case. Then we try to give a complete picture of the existing applications and we present some related open problems. | [
{
"version": "v1",
"created": "Mon, 10 Feb 2025 11:12:43 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 12:01:04 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Fuhrman",
"Marco",
""
]
] |
2502.06965 | Lillian Luo | Lillian Luo and Maxim Perelstein | Conformal Freeze-in Dark Matter: 5D Dual and Phase Transition | 21 pages, 7 figures. Version accepted for publication in JHEP | null | null | null | hep-ph | http://creativecommons.org/licenses/by/4.0/ | Conformal Freeze-in (COFI) scenario postulates a dark sector described by a conformal field theory (CFT) at energies above the ``gap scale" in the keV$-$MeV range. At the gap scale, the dark CFT undergoes confinement, and one of the resulting bound states is identified as the dark matter candidate. In this paper, we study this model in the context of the AdS/CFT correspondence with a focus on the mechanism of the infrared (IR) breaking of conformal invariance in the dark sector. We construct the holographic dual to the conformal dark sector, given by a Randall-Sundrum-like model in 5D, where the Standard Model (SM) fields and the dark matter candidate are placed on the ultraviolet (UV) and IR branes respectively. The separation between the UV and IR branes is stabilized by a bulk scalar field, naturally generating a hierarchy between the electroweak scale and the gap scale. We find that the parameter space of COFI comprises two distinct branches of CFT's living on the Anti-de-Sitter (AdS) boundary, each corresponding to a different UV boundary condition. The two branches of CFT's result in different radion potentials. The confinement of the CFT is dual to the spontaneous symmetry breaking by the 5D radion potential. We then use this dual 5D setup to study the cosmological confining phase transition in the dark sector. We find the viable parameter space of the theory which allows the phase transition to complete promptly without significant supercooling. | [
{
"version": "v1",
"created": "Mon, 10 Feb 2025 19:02:33 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 17:51:57 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Luo",
"Lillian",
""
],
[
"Perelstein",
"Maxim",
""
]
] |
2502.07202 | Jaesik Yoon | Jaesik Yoon, Hyeonseo Cho, Doojin Baek, Yoshua Bengio and Sungjin Ahn | Monte Carlo Tree Diffusion for System 2 Planning | 23 pages, 7 figures, ICML 2025 Main Track Spotlight | null | null | null | cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Diffusion models have recently emerged as a powerful tool for planning. However, unlike Monte Carlo Tree Search (MCTS)-whose performance naturally improves with inference-time computation scaling-standard diffusion-based planners offer only limited avenues for the scalability. In this paper, we introduce Monte Carlo Tree Diffusion (MCTD), a novel framework that integrates the generative strength of diffusion models with the adaptive search capabilities of MCTS. Our method reconceptualizes denoising as a tree-structured process, allowing partially denoised plans to be iteratively evaluated, pruned, and refined. By selectively expanding promising trajectories while retaining the flexibility to revisit and improve suboptimal branches, MCTD achieves the benefits of MCTS such as controlling exploration-exploitation trade-offs within the diffusion framework. Empirical results on challenging long-horizon tasks show that MCTD outperforms diffusion baselines, yielding higher-quality solutions as inference-time computation increases. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 02:51:42 GMT"
},
{
"version": "v2",
"created": "Fri, 11 Apr 2025 00:14:32 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 00:37:56 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 01:22:31 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Yoon",
"Jaesik",
""
],
[
"Cho",
"Hyeonseo",
""
],
[
"Baek",
"Doojin",
""
],
[
"Bengio",
"Yoshua",
""
],
[
"Ahn",
"Sungjin",
""
]
] |
2502.07252 | Ehsan Khatami | Robin C. Newby and Ehsan Khatami | Finite-Temperature Kinetic Ferromagnetism in the Square Lattice Hubbard Model | 11 pages, 10 figures | Physical Review B 111, 245120 (2025) | 10.1103/PhysRevB.111.245120 | null | cond-mat.str-el cond-mat.quant-gas | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | While the exact phase diagram of the Fermi-Hubbard model remains poorly understood despite decades of progress, nearly 60 years ago, Nagaoka proved that a single dopant in an otherwise half-filled Hubbard system can bring about ferromagnetism through kinetic means. The phenomenon was recently observed with ultracold atoms in triangular optical lattices. Here, we explore the kinetic ferromagnetism within the square lattice Hubbard model and its strong-coupling counterpart, the $t-J$ model, at finite temperatures in the thermodynamic limit via numerical linked-cluster expansions. We find evidence of ferromagnetic Nagaoka polarons at dopings up to $\sim 30\%$ away from half filling for a variety of interaction strengths and at temperatures as low as $0.2$ of the hopping energy. We map out the boundaries of this phase through analyzing various correlation functions. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 04:37:46 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:33:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Newby",
"Robin C.",
""
],
[
"Khatami",
"Ehsan",
""
]
] |
2502.07783 | Leyang Hu | Leyang Hu, Matteo Gamba, Randall Balestriero | Curvature Tuning: Provable Training-free Model Steering From a Single Parameter | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | The scaling of model and data sizes has reshaped the AI landscape, establishing finetuning pretrained models as the standard paradigm for solving downstream tasks. However, dominant finetuning methods typically rely on weight adaptation, often lack interpretability, and depend on heuristically chosen hyperparameters. In this paper, we take a different perspective and shift the focus from weights to activation functions, viewing them through the lens of spline operators. We propose Curvature Tuning (CT), an interpretable and principled steering method that modulates a model's decision boundary by injecting a single hyperparameter into its activation functions. We show that CT provably adjusts model decision boundary curvature and, more fundamentally, projects a model onto a space of smooth functions-thereby complementing current finetuning methods, whose effect lies primarily in feature adaptation. Making this hyperparameter trainable gives rise to a novel and highly parameter-efficient finetuning method. Empirically, CT improves both generalization and robustness. For example, it boosts downstream accuracy of ResNet-50/152 by 7.14%/8.46% over linear probing and 4.64%/1.70% over LoRA across 12 datasets, and improves robust accuracy on the $\ell_\infty$ benchmark from RobustBench by 1032.64%/1494.46%. Our code is available at https://github.com/Leon-Leyang/curvature-tuning. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 18:59:57 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 17:59:08 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 17:37:05 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 17:09:08 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Hu",
"Leyang",
""
],
[
"Gamba",
"Matteo",
""
],
[
"Balestriero",
"Randall",
""
]
] |
2502.08014 | Nutth Tuchinda | Nutth Tuchinda, Changle Li, Christopher A. Schuh | The Augmented Potential Method: Multiscale Modeling Toward a Spectral Defect Genome | null | null | null | null | cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The modeling of solute chemistry at low-symmetry defects in materials is historically challenging, due to the computation cost required to evaluate thermodynamic properties from first principles. Here, we offer a hybrid multiscale approach called the augmented potential method that connects the chemical flexibility and near-quantum accuracy of a universal machine learning potential at the site of the defect, with the computational speed of a long-range classical potential implemented away from the defect site in a buffer zone. The method allows us to rapidly compute distributions of grain boundary segregation energy for 1,050 binary alloy pairs (including Ag, Al, Au, Cr, Cu, Fe, Mo, Nb, Ni, Pd, Pt, Ta and V, W solvent), creating a database for polycrystalline grain boundary segregation. This database is ~5x larger than previously published spectral compilations, and yet has improved accuracy. The approach can also address problems far beyond the reach of any other method, such as handling bcc Fe-based alloys, or the complex solute-solute interactions in random polycrystals. The approach thus paves a pathway toward a complete defect genome in crystalline materials. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 23:26:46 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 12:25:06 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Tuchinda",
"Nutth",
""
],
[
"Li",
"Changle",
""
],
[
"Schuh",
"Christopher A.",
""
]
] |
2502.09101 | Feihong Lu | Zongyu Chang, Feihong Lu, Ziqin Zhu, Qian Li, Cheng Ji, Zhuo Chen, Hao Peng, Yang Liu, Ruifeng Xu, Yangqiu Song, Shangguang Wang, Jianxin Li | Bridging the Gap Between LLMs and Human Intentions: Progresses and Challenges in Instruction Understanding, Intention Reasoning, and Reliable Generation | 19 pages, 11 figures | null | null | null | cs.HC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large language models (LLMs) have demonstrated exceptional capabilities in understanding and generation. However, when interacting with human instructions in real-world scenarios, LLMs still face significant challenges, particularly in accurately capturing and comprehending human instructions and intentions. This paper focuses on three challenges in LLM-based text generation tasks: instruction understanding, intention reasoning, and Reliable Dialog Generation. Regarding human complex instruction, LLMs have deficiencies in understanding long contexts and instructions in multi-round conversations. For intention reasoning, LLMs may have inconsistent command reasoning, difficulty reasoning about commands containing incorrect information, difficulty understanding user ambiguous language commands, and a weak understanding of user intention in commands. Besides, In terms of Reliable Dialog Generation, LLMs may have unstable generated content and unethical generation. To this end, we classify and analyze the performance of LLMs in challenging scenarios and conduct a comprehensive evaluation of existing solutions. Furthermore, we introduce benchmarks and categorize them based on the aforementioned three core challenges. Finally, we explore potential directions for future research to enhance the reliability and adaptability of LLMs in real-world applications. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 09:19:42 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 04:23:09 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Chang",
"Zongyu",
""
],
[
"Lu",
"Feihong",
""
],
[
"Zhu",
"Ziqin",
""
],
[
"Li",
"Qian",
""
],
[
"Ji",
"Cheng",
""
],
[
"Chen",
"Zhuo",
""
],
[
"Peng",
"Hao",
""
],
[
"Liu",
"Yang",
""
],
[
"Xu",
"Ruifeng",
""
],
[
"Song",
"Yangqiu",
""
],
[
"Wang",
"Shangguang",
""
],
[
"Li",
"Jianxin",
""
]
] |
2502.09252 | Andrew Draganov | Andrew Draganov, Sharvaree Vadgama, Sebastian Damrich, Jan Niklas B\"ohm, Lucas Maes, Dmitry Kobak and Erik Bekkers | On the Importance of Embedding Norms in Self-Supervised Learning | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Self-supervised learning (SSL) allows training data representations without a supervised signal and has become an important paradigm in machine learning. Most SSL methods employ the cosine similarity between embedding vectors and hence effectively embed data on a hypersphere. While this seemingly implies that embedding norms cannot play any role in SSL, a few recent works have suggested that embedding norms have properties related to network convergence and confidence. In this paper, we resolve this apparent contradiction and systematically establish the embedding norm's role in SSL training. Using theoretical analysis, simulations, and experiments, we show that embedding norms (i) govern SSL convergence rates and (ii) encode network confidence, with smaller norms corresponding to unexpected samples. Additionally, we show that manipulating embedding norms can have large effects on convergence speed. Our findings demonstrate that SSL embedding norms are integral to understanding and optimizing network behavior. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 12:09:17 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:17:56 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Draganov",
"Andrew",
""
],
[
"Vadgama",
"Sharvaree",
""
],
[
"Damrich",
"Sebastian",
""
],
[
"Böhm",
"Jan Niklas",
""
],
[
"Maes",
"Lucas",
""
],
[
"Kobak",
"Dmitry",
""
],
[
"Bekkers",
"Erik",
""
]
] |
2502.09330 | Raghunathan Ramakrishnan Prof. | Atreyee Majumdar and Raghunathan Ramakrishnan | Leveraging the Bias-Variance Tradeoff in Quantum Chemistry for Accurate Negative Singlet-Triplet Gap Predictions: A Case for Double-Hybrid DFT | Please download version 2 to access SI | null | null | null | physics.chem-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Molecules that have been suggested to violate the Hund's rule, having a first excited singlet state (S$_1$) energetically below the triplet state (T$_1$), are rare. Yet, they hold the promise to be efficient light emitters. Their high-throughput identification demands exceptionally accurate excited-state modeling to minimize qualitatively wrong predictions. We benchmark twelve S$_1$-T$_1$ energy gaps to find that the local-correlated versions of ADC(2) and CC2 excited state methods deliver excellent accuracy and speed for screening medium-sized molecules. Notably, we find that double-hybrid DFT approximations (e.g., B2GP-PLYP and PBE-QIDH) exhibit high mean absolute errors ($>100$ meV) despite very low standard deviations ($\approx10$ meV). Exploring their parameter space reveals that a configuration with 75% exchange and 55% correlation, which reduces the mean absolute error to below 5 meV, but with an increased variance. Using this low-bias parameterization as an internal reference, we correct the systematic error while maintaining low variance, effectively combining the strengths of both low-bias and low-variance DFT parameterizations to enhance overall accuracy. Our findings suggest that low-variance DFT methods, often overlooked due to their high bias, can serve as reliable tools for predictive modeling in first-principles molecular design. The bias-correction data-fitting procedure can be applied to any general problem where two flavors of a method, one with low bias and another with low variance, have been identified a priori. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 13:47:16 GMT"
},
{
"version": "v2",
"created": "Fri, 14 Feb 2025 06:38:58 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 10:12:02 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 07:08:23 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Majumdar",
"Atreyee",
""
],
[
"Ramakrishnan",
"Raghunathan",
""
]
] |
2502.09502 | Jiachang Liu | Jiachang Liu, Soroosh Shafiee, Andrea Lodi | Scalable First-order Method for Certifying Optimal k-Sparse GLMs | ICML 2025 camera ready, typo fixed | null | null | null | cs.LG math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper investigates the problem of certifying optimality for sparse generalized linear models (GLMs), where sparsity is enforced through an $\ell_0$ cardinality constraint. While branch-and-bound (BnB) frameworks can certify optimality by pruning nodes using dual bounds, existing methods for computing these bounds are either computationally intensive or exhibit slow convergence, limiting their scalability to large-scale problems. To address this challenge, we propose a first-order proximal gradient algorithm designed to solve the perspective relaxation of the problem within a BnB framework. Specifically, we formulate the relaxed problem as a composite optimization problem and demonstrate that the proximal operator of the non-smooth component can be computed exactly in log-linear time complexity, eliminating the need to solve a computationally expensive second-order cone program. Furthermore, we introduce a simple restart strategy that enhances convergence speed while maintaining low per-iteration complexity. Extensive experiments on synthetic and real-world datasets show that our approach significantly accelerates dual bound computations and is highly effective in providing optimality certificates for large-scale problems. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 17:14:18 GMT"
},
{
"version": "v2",
"created": "Thu, 05 Jun 2025 18:08:53 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 17:23:10 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Liu",
"Jiachang",
""
],
[
"Shafiee",
"Soroosh",
""
],
[
"Lodi",
"Andrea",
""
]
] |
2502.09720 | Semyon Savkin | Semyon Savkin, Eitan Porat, Or Ordentlich, Yury Polyanskiy | NestQuant: Nested Lattice Quantization for Matrix Products and LLMs | 23 pages | null | null | null | cs.LG cs.AI cs.IT math.IT | http://creativecommons.org/licenses/by/4.0/ | Post-training quantization (PTQ) has emerged as a critical technique for efficient deployment of large language models (LLMs). This work proposes NestQuant, a novel PTQ scheme for weights and activations that is based on self-similar nested lattices. Recent works have mathematically shown such quantizers to be information-theoretically optimal for low-precision matrix multiplication. We implement a practical low-complexity version of NestQuant based on Gosset lattice, making it a drop-in quantizer for any matrix multiplication step (e.g., in self-attention, MLP etc). For example, NestQuant quantizes weights, KV-cache, and activations of Llama-3-8B to 4 bits, achieving perplexity of 6.6 on wikitext2. This represents more than 55% reduction in perplexity gap with respect to unquantized model (perplexity of 6.14) compared to state-of-the-art Metas SpinQuant (perplexity 7.3), OstQuant (7.3) and QuaRot (8.2). Comparisons on bigger models (up to 70B) and on various LLM evaluation benchmarks confirm uniform superiority of NestQuant. | [
{
"version": "v1",
"created": "Thu, 13 Feb 2025 19:11:40 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 06:01:15 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Savkin",
"Semyon",
""
],
[
"Porat",
"Eitan",
""
],
[
"Ordentlich",
"Or",
""
],
[
"Polyanskiy",
"Yury",
""
]
] |
2502.10104 | Sofie Ried | Steffen Gielen and Sofie Ried | Quantum Schwarzschild-(A)dS Black Holes: Unitarity and Singularity Resolution | minor changes to the presentation, accepted for publication in JHEP, 12 figures and 47 pages | JHEP 06 (2025) 074 | 10.1007/JHEP06(2025)074 | null | hep-th gr-qc | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We consider the canonical quantisation of spherically symmetric spacetimes within unimodular gravity, leaving sign choices in the metric general enough to include both the interior and exterior Schwarzschild-(Anti-)de Sitter spacetime. In unimodular gravity the cosmological constant appears as an integration constant analogous to a total energy, and the quantum Wheeler-DeWitt equation takes the form of a Schr\"odinger equation in unimodular time. We discuss self-adjoint extensions of the Schr\"odinger-like Hamiltonian arising from the requirement of unitarity in unimodular time, and identify a physically motivated one-parameter family of extensions. For semiclassical states we are able to derive analytical expressions for expectation values of the metric, representing a quantum-corrected, nonsingular extension of the classical Schwarzschild-(A)dS geometry which describes a quantum transition between asymptotic black hole and white hole states. The sign of the self-adjoint extension parameter corresponds to the allowed sign of the black hole/white hole mass, and so it can be chosen to ensure that this mass is always positive. We also discuss tunnelling states which allow for a change in the sign of the mass, but which are not semiclassical in high-curvature regions. Our mechanism for singularity resolution and the explicit form of the quantum-corrected metric can be compared to other proposals for black holes in quantum gravity, and in the asymptotically AdS case can be contrasted with holographic arguments. | [
{
"version": "v1",
"created": "Fri, 14 Feb 2025 11:59:15 GMT"
},
{
"version": "v2",
"created": "Fri, 09 May 2025 15:08:25 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Gielen",
"Steffen",
""
],
[
"Ried",
"Sofie",
""
]
] |
2502.10450 | Jung-Eun Kim | Xingli Fang, Jianwei Li, Varun Mulchandani, Jung-Eun Kim | Trustworthy AI: Safety, Bias, and Privacy -- A Survey | null | null | null | null | cs.CR cs.AI cs.CL cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | The capabilities of artificial intelligence systems have been advancing to a great extent, but these systems still struggle with failure modes, vulnerabilities, and biases. In this paper, we study the current state of the field, and present promising insights and perspectives regarding concerns that challenge the trustworthiness of AI models. In particular, this paper investigates the issues regarding three thrusts: safety, privacy, and bias, which hurt models' trustworthiness. For safety, we discuss safety alignment in the context of large language models, preventing them from generating toxic or harmful content. For bias, we focus on spurious biases that can mislead a network. Lastly, for privacy, we cover membership inference attacks in deep neural networks. The discussions addressed in this paper reflect our own experiments and observations. | [
{
"version": "v1",
"created": "Tue, 11 Feb 2025 20:08:42 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 17:44:34 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Fang",
"Xingli",
""
],
[
"Li",
"Jianwei",
""
],
[
"Mulchandani",
"Varun",
""
],
[
"Kim",
"Jung-Eun",
""
]
] |
2502.11420 | Yingqing Guo | Yingqing Guo, Yukang Yang, Hui Yuan, Mengdi Wang | Training-Free Guidance Beyond Differentiability: Scalable Path Steering with Tree Search in Diffusion and Flow Models | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Training-free guidance enables controlled generation in diffusion and flow models, but most methods rely on gradients and assume differentiable objectives. This work focuses on training-free guidance addressing challenges from non-differentiable objectives and discrete data distributions. We propose TreeG: Tree Search-Based Path Steering Guidance, applicable to both continuous and discrete settings in diffusion and flow models. TreeG offers a unified framework for training-free guidance by proposing, evaluating, and selecting candidates at each step, enhanced with tree search over active paths and parallel exploration. We comprehensively investigate the design space of TreeG over the candidate proposal module and the evaluation function, instantiating TreeG into three novel algorithms. Our experiments show that TreeG consistently outperforms top guidance baselines in symbolic music generation, small molecule design, and enhancer DNA design with improvements of 29.01%, 16.6%, and 18.43%. Additionally, we identify an inference-time scaling law showing TreeG's scalability in inference-time computation. | [
{
"version": "v1",
"created": "Mon, 17 Feb 2025 04:20:39 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 01:06:22 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 01:18:01 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Guo",
"Yingqing",
""
],
[
"Yang",
"Yukang",
""
],
[
"Yuan",
"Hui",
""
],
[
"Wang",
"Mengdi",
""
]
] |
2502.11940 | Vincenzo Petrone | Vincenzo Petrone, Enrico Ferrentino, and Pasquale Chiacchio | The Dynamic Model of the UR10 Robot and its ROS2 Integration | 11 pages, 6 figures, 6 tables, IEEE Transactions on Industrial Informatics | EEE Transactions on Industrial Informatics, vol. 21, no. 5, pp. 3828-3838, May 2025 | 10.1109/TII.2025.3534415 | null | cs.RO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper presents the full dynamic model of the UR10 industrial robot. A triple-stage identification approach is adopted to estimate the manipulator's dynamic coefficients. First, linear parameters are computed using a standard linear regression algorithm. Subsequently, nonlinear friction parameters are estimated according to a sigmoidal model. Lastly, motor drive gains are devised to map estimated joint currents to torques. The overall identified model can be used for both control and planning purposes, as the accompanied ROS2 software can be easily reconfigured to account for a generic payload. The estimated robot model is experimentally validated against a set of exciting trajectories and compared to the state-of-the-art model for the same manipulator, achieving higher current prediction accuracy (up to a factor of 4.43) and more precise motor gains. The related software is available at https://codeocean.com/capsule/8515919/tree/v2. | [
{
"version": "v1",
"created": "Mon, 17 Feb 2025 15:51:57 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Petrone",
"Vincenzo",
""
],
[
"Ferrentino",
"Enrico",
""
],
[
"Chiacchio",
"Pasquale",
""
]
] |
2502.12112 | Keisuke Inomata | Keisuke Inomata | Role of the counterterms in the conservation of superhorizon curvature perturbations at one loop | 7 pages, v3: published version | null | 10.1103/r8bh-s48f | null | astro-ph.CO gr-qc hep-ph hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recently, several papers have claimed that superhorizon curvature perturbations are not conserved at the one-loop level in single-field inflation models if there is a transient ultra-slow-roll period. In this work, we point out that the contributions from the counterterms were overlooked in the recent papers. We show that the counterterm contributions play a crucial role in canceling the one-loop power spectrum of superhorizon curvature perturbations in the comoving gauge. | [
{
"version": "v1",
"created": "Mon, 17 Feb 2025 18:36:33 GMT"
},
{
"version": "v2",
"created": "Tue, 25 Feb 2025 12:28:16 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 17:33:35 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Inomata",
"Keisuke",
""
]
] |
2502.12268 | Laura Monk | Nalini Anantharaman and Laura Monk | Friedman-Ramanujan functions in random hyperbolic geometry and application to spectral gaps II | 160 pages, 43 figures. Comments welcome! | null | null | null | math.MG math.SP | http://creativecommons.org/licenses/by/4.0/ | The core focus of this series of two articles is the study of the distribution of the length spectrum of closed hyperbolic surfaces of genus $g$, sampled randomly with respect to the Weil-Petersson probability measure. In the first article, we introduced a notion of local topological type $T$, and established the existence of a density function $V_g^T(l)$ describing the distribution of the lengths of all closed geodesics of type $T$ in a genus $g$ hyperbolic surface. We proved that $V_g^{T}(l)$ admits an asymptotic expansion in powers of $1/g$. We introduced a new class of functions, called Friedman-Ramanujan functions, and related it to the study of the spectral gap $\lambda_1$ of the Laplacian.
In this second part, we provide a variety of new tools allowing to compute and estimate the volume functions $V_g^{T}(l)$. Notably, we construct new sets of coordinates on Teichm\"uller spaces, distinct from Fenchel-Nielsen coordinates, in which the Weil-Petersson volume has a simple form. These coordinates are tailored to the geodesics we study, and we can therefore prove nice formulae for their lengths. We use these new ideas, together with a notion of pseudo-convolutions, to prove that the coefficients of the expansion of $V_g^{T}(l)$ in powers of $1/g$ are Friedman-Ramanujan functions, for any local topological type $T$. We then exploit this result to prove that, for any $\epsilon>0$, $\lambda_1 \geq \frac14 - \epsilon$ with probability going to one as $g \rightarrow + \infty$, or, in other words, typical hyperbolic surfaces have an asymptotically optimal spectral gap. | [
{
"version": "v1",
"created": "Mon, 17 Feb 2025 19:08:30 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:06:59 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Anantharaman",
"Nalini",
""
],
[
"Monk",
"Laura",
""
]
] |
2502.13131 | Feng Luo | Feng Luo, Rui Yang, Hao Sun, Chunyuan Deng, Jiarui Yao, Jingyan Shen, Huan Zhang, Hanjie Chen | Rethinking Diverse Human Preference Learning through Principal Component Analysis | 14 pages | null | null | null | cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Understanding human preferences is crucial for improving foundation models and building personalized AI systems. However, preferences are inherently diverse and complex, making it difficult for traditional reward models to capture their full range. While fine-grained preference data can help, collecting it is expensive and hard to scale. In this paper, we introduce Decomposed Reward Models (DRMs), a novel approach that extracts diverse human preferences from binary comparisons without requiring fine-grained annotations. Our key insight is to represent human preferences as vectors and analyze them using Principal Component Analysis (PCA). By constructing a dataset of embedding differences between preferred and rejected responses, DRMs identify orthogonal basis vectors that capture distinct aspects of preference. These decomposed rewards can be flexibly combined to align with different user needs, offering an interpretable and scalable alternative to traditional reward models. We demonstrate that DRMs effectively extract meaningful preference dimensions (e.g., helpfulness, safety, humor) and adapt to new users without additional training. Our results highlight DRMs as a powerful framework for personalized and interpretable LLM alignment. Our code is available at https://github.com/amandaluof/DRMs. | [
{
"version": "v1",
"created": "Tue, 18 Feb 2025 18:55:26 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 05:43:13 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Luo",
"Feng",
""
],
[
"Yang",
"Rui",
""
],
[
"Sun",
"Hao",
""
],
[
"Deng",
"Chunyuan",
""
],
[
"Yao",
"Jiarui",
""
],
[
"Shen",
"Jingyan",
""
],
[
"Zhang",
"Huan",
""
],
[
"Chen",
"Hanjie",
""
]
] |
2502.13191 | Junyi Guan | Junyi Guan, Abhijith Sharma, Chong Tian, Salem Lahlou | On the Privacy Risks of Spiking Neural Networks: A Membership Inference Analysis | 14 pages, 6 figures | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Spiking Neural Networks (SNNs) are increasingly explored for their energy efficiency and robustness in real-world applications, yet their privacy risks remain largely unexamined. In this work, we investigate the susceptibility of SNNs to Membership Inference Attacks (MIAs) -- a major privacy threat where an adversary attempts to determine whether a given sample was part of the training dataset. While prior work suggests that SNNs may offer inherent robustness due to their discrete, event-driven nature, we find that its resilience diminishes as latency (T) increases. Furthermore, we introduce an input dropout strategy under black box setting, that significantly enhances membership inference in SNNs. Our findings challenge the assumption that SNNs are inherently more secure, and even though they are expected to be better, our results reveal that SNNs exhibit privacy vulnerabilities that are equally comparable to Artificial Neural Networks (ANNs). Our code is available at https://github.com/sharmaabhijith/MIA_SNN. | [
{
"version": "v1",
"created": "Tue, 18 Feb 2025 15:19:20 GMT"
},
{
"version": "v2",
"created": "Sun, 16 Mar 2025 15:25:29 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 10:38:35 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 09:08:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Guan",
"Junyi",
""
],
[
"Sharma",
"Abhijith",
""
],
[
"Tian",
"Chong",
""
],
[
"Lahlou",
"Salem",
""
]
] |
2502.13228 | Jake Snell | Jake C. Snell, Thomas L. Griffiths | Conformal Prediction as Bayesian Quadrature | ICML 2025 camera-ready version (accepted as an oral presentation). 16 pages, 4 figures. Code available at https://github.com/jakesnell/conformal-as-bayes-quad | null | null | null | cs.LG cs.AI stat.ML | http://creativecommons.org/licenses/by/4.0/ | As machine learning-based prediction systems are increasingly used in high-stakes situations, it is important to understand how such predictive models will perform upon deployment. Distribution-free uncertainty quantification techniques such as conformal prediction provide guarantees about the loss black-box models will incur even when the details of the models are hidden. However, such methods are based on frequentist probability, which unduly limits their applicability. We revisit the central aspects of conformal prediction from a Bayesian perspective and thereby illuminate the shortcomings of frequentist guarantees. We propose a practical alternative based on Bayesian quadrature that provides interpretable guarantees and offers a richer representation of the likely range of losses to be observed at test time. | [
{
"version": "v1",
"created": "Tue, 18 Feb 2025 19:06:21 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:39:13 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Snell",
"Jake C.",
""
],
[
"Griffiths",
"Thomas L.",
""
]
] |
2502.13338 | Peter R. Wiecha | Dalin Soun and Antoine Az\'ema and Lucien Roach and Glenna L. Drisko and Peter R. Wiecha | Gradient-Based Optimization of Core-Shell Particles with Discrete Materials for Directional Scattering | 7 pages, 7 figures; sup.info pdf of 6 pages | Optics Express 33(12), 25945-25958 (2025) | 10.1364/OE.561657 | null | physics.optics physics.comp-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Designing nanophotonic structures traditionally grapples with the complexities of discrete parameters, such as real materials, often resorting to costly global optimization methods. This paper introduces an approach that leverages generative deep learning to map discrete parameter sets into a continuous latent space, enabling direct gradient-based optimization. For scenarios with non-differentiable physics evaluation functions, a neural network is employed as a differentiable surrogate model. The efficacy of this methodology is demonstrated by optimizing the directional scattering properties of core-shell nanoparticles composed of a selection of realistic materials. We derive suggestions for core-shell geometries with strong forward scattering and minimized backscattering. Our findings reveal significant improvements in computational efficiency and performance when compared to global optimization techniques. Beyond nanophotonics design problems, this framework holds promise for broad applications across all types of inverse problems constrained by discrete variables. | [
{
"version": "v1",
"created": "Tue, 18 Feb 2025 23:38:25 GMT"
},
{
"version": "v2",
"created": "Tue, 20 May 2025 10:53:24 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Soun",
"Dalin",
""
],
[
"Azéma",
"Antoine",
""
],
[
"Roach",
"Lucien",
""
],
[
"Drisko",
"Glenna L.",
""
],
[
"Wiecha",
"Peter R.",
""
]
] |
2502.13667 | Leon Chini | Leon Chini | Model Theory of Generic Vector Space Endomorphisms | 40 pages | null | null | null | math.LO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper deals with the model companion of an endomorphism acting on a vector space, possibly with extra structure. Given a theory $T$ that $\varnothing$-defines an infinite $K$-vector space $\mathbb{V}$ in every model, we set $T_\theta := T \cup \{\text{``$\theta$ defines a $K$-endomorphism of $\mathbb{V}$"}\}$. We then consider extensions of the form $$ T_\theta \cup \left\{\sum\nolimits_{k}\bigcap\nolimits_{l}\operatorname{Ker}(\rho_{j, k, l}[\theta]) = \sum\nolimits_{k}\bigcap\nolimits_{l} \operatorname{Ker}(\eta_{j, k, l}[\theta]) : j \in \mathcal{J}\right\}, $$ where all sums and intersections are finite, and all the $\rho[\theta]$'s and $\eta[\theta]$'s are polynomials over $K$ with $\theta$ plugged in. Notice that properties such as $\theta^2 - 2\operatorname{Id} = 0$ or $\operatorname{Ker}(\theta^n) = \operatorname{Ker}(\theta^{n+1})$ can be expressed in such a manner. We then parametrize the consistent extensions of this form by a family $\{T^C_\theta : C \in \mathcal{C}\}$ and characterize the existentially closed models of each $T^C_\theta$. We also present a sufficient criterion, which only depends on $T$, for when these characterizations are first-order expressible, i.e., for when a model companion of each $T^C_\theta$ exists. | [
{
"version": "v1",
"created": "Wed, 19 Feb 2025 12:23:18 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 20:54:53 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Chini",
"Leon",
""
]
] |
2502.13909 | Sein Kim | Sein Kim, Hongseok Kang, Kibum Kim, Jiwan Kim, Donghyun Kim, Minchul Yang, Kwangjin Oh, Julian McAuley, Chanyoung Park | Lost in Sequence: Do Large Language Models Understand Sequential Recommendation? | KDD 2025 Research Track | null | null | null | cs.IR cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Language Models (LLMs) have recently emerged as promising tools for recommendation thanks to their advanced textual understanding ability and context-awareness. Despite the current practice of training and evaluating LLM-based recommendation (LLM4Rec) models under a sequential recommendation scenario, we found that whether these models understand the sequential information inherent in users' item interaction sequences has been largely overlooked. In this paper, we first demonstrate through a series of experiments that existing LLM4Rec models do not fully capture sequential information both during training and inference. Then, we propose a simple yet effective LLM-based sequential recommender, called LLM-SRec, a method that enhances the integration of sequential information into LLMs by distilling the user representations extracted from a pre-trained CF-SRec model into LLMs. Our extensive experiments show that LLM-SRec enhances LLMs' ability to understand users' item interaction sequences, ultimately leading to improved recommendation performance. Furthermore, unlike existing LLM4Rec models that require fine-tuning of LLMs, LLM-SRec achieves state-of-the-art performance by training only a few lightweight MLPs, highlighting its practicality in real-world applications. Our code is available at https://github.com/Sein-Kim/LLM-SRec. | [
{
"version": "v1",
"created": "Wed, 19 Feb 2025 17:41:09 GMT"
},
{
"version": "v2",
"created": "Fri, 21 Feb 2025 03:12:46 GMT"
},
{
"version": "v3",
"created": "Wed, 02 Apr 2025 17:42:03 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 17:41:16 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Kim",
"Sein",
""
],
[
"Kang",
"Hongseok",
""
],
[
"Kim",
"Kibum",
""
],
[
"Kim",
"Jiwan",
""
],
[
"Kim",
"Donghyun",
""
],
[
"Yang",
"Minchul",
""
],
[
"Oh",
"Kwangjin",
""
],
[
"McAuley",
"Julian",
""
],
[
"Park",
"Chanyoung",
""
]
] |
2502.13961 | Yatin Dandi | Yatin Dandi, Luca Pesce, Lenka Zdeborov\'a, Florent Krzakala | The Computational Advantage of Depth: Learning High-Dimensional Hierarchical Functions with Gradient Descent | null | null | null | null | stat.ML cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Understanding the advantages of deep neural networks trained by gradient descent (GD) compared to shallow models remains an open theoretical challenge. In this paper, we introduce a class of target functions (single and multi-index Gaussian hierarchical targets) that incorporate a hierarchy of latent subspace dimensionalities. This framework enables us to analytically study the learning dynamics and generalization performance of deep networks compared to shallow ones in the high-dimensional limit. Specifically, our main theorem shows that feature learning with GD successively reduces the effective dimensionality, transforming a high-dimensional problem into a sequence of lower-dimensional ones. This enables learning the target function with drastically less samples than with shallow networks. While the results are proven in a controlled training setting, we also discuss more common training procedures and argue that they learn through the same mechanisms. | [
{
"version": "v1",
"created": "Wed, 19 Feb 2025 18:58:28 GMT"
},
{
"version": "v2",
"created": "Mon, 09 Jun 2025 22:34:01 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 04:59:36 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Dandi",
"Yatin",
""
],
[
"Pesce",
"Luca",
""
],
[
"Zdeborová",
"Lenka",
""
],
[
"Krzakala",
"Florent",
""
]
] |
2502.14254 | Lingfeng Zhang | Lingfeng Zhang, Yuecheng Liu, Zhanguang Zhang, Matin Aghaei, Yaochen Hu, Hongjian Gu, Mohammad Ali Alomrani, David Gamaliel Arcos Bravo, Raika Karimi, Atia Hamidizadeh, Haoping Xu, Guowei Huang, Zhanpeng Zhang, Tongtong Cao, Weichao Qiu, Xingyue Quan, Jianye Hao, Yuzheng Zhuang, Yingxue Zhang | Mem2Ego: Empowering Vision-Language Models with Global-to-Ego Memory for Long-Horizon Embodied Navigation | null | null | null | null | cs.RO cs.AI | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Recent advancements in Large Language Models (LLMs) and Vision-Language Models (VLMs) have made them powerful tools in embodied navigation, enabling agents to leverage commonsense and spatial reasoning for efficient exploration in unfamiliar environments. Existing LLM-based approaches convert global memory, such as semantic or topological maps, into language descriptions to guide navigation. While this improves efficiency and reduces redundant exploration, the loss of geometric information in language-based representations hinders spatial reasoning, especially in intricate environments. To address this, VLM-based approaches directly process ego-centric visual inputs to select optimal directions for exploration. However, relying solely on a first-person perspective makes navigation a partially observed decision-making problem, leading to suboptimal decisions in complex environments. In this paper, we present a novel vision-language model (VLM)-based navigation framework that addresses these challenges by adaptively retrieving task-relevant cues from a global memory module and integrating them with the agent's egocentric observations. By dynamically aligning global contextual information with local perception, our approach enhances spatial reasoning and decision-making in long-horizon tasks. Experimental results demonstrate that the proposed method surpasses previous state-of-the-art approaches in object navigation tasks, providing a more effective and scalable solution for embodied navigation. | [
{
"version": "v1",
"created": "Thu, 20 Feb 2025 04:41:40 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 21:36:52 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Zhang",
"Lingfeng",
""
],
[
"Liu",
"Yuecheng",
""
],
[
"Zhang",
"Zhanguang",
""
],
[
"Aghaei",
"Matin",
""
],
[
"Hu",
"Yaochen",
""
],
[
"Gu",
"Hongjian",
""
],
[
"Alomrani",
"Mohammad Ali",
""
],
[
"Bravo",
"David Gamaliel Arcos",
""
],
[
"Karimi",
"Raika",
""
],
[
"Hamidizadeh",
"Atia",
""
],
[
"Xu",
"Haoping",
""
],
[
"Huang",
"Guowei",
""
],
[
"Zhang",
"Zhanpeng",
""
],
[
"Cao",
"Tongtong",
""
],
[
"Qiu",
"Weichao",
""
],
[
"Quan",
"Xingyue",
""
],
[
"Hao",
"Jianye",
""
],
[
"Zhuang",
"Yuzheng",
""
],
[
"Zhang",
"Yingxue",
""
]
] |
2502.14608 | Cl\'ement Stahl | Cl\'ement Stahl, Vivian Poulin, Benoit Famaey, Rodrigo Ibata | A primordial origin to cosmic tensions: towards reconciling $H_0$ and $S_8$ with early dark energy and scale-dependent primordial non-Gaussianities | are still welcome :) 17 pages, 3 figures, 2 tables, JCAP accepted | null | null | null | astro-ph.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The Hubble ($H_0$) tension between direct measurements of the expansion rate and the prediction of the $\Lambda$CDM cosmological model calibrated on the Cosmic Microwave Background (CMB), is a strong motivation to explore alternative cosmological models. A popular class of such models includes an additional component of dark energy relevant in the early Universe, which solves the Hubble tension. These Early Dark Energy (EDE) models however typically overpredict the value of the $S_8$ parameter. Here, we show how combining EDE with scale-dependent primordial non-Gaussianities (sPNG) can in principle both increase $H_0$ and decrease $S_8$ at once, even conceivably allowing to solve the potential $S_8$ tension between measurements of weak gravitational lensing and the $\Lambda$CDM expectation. Such sPNG are related to non-trivial inflationary physics, and observationally affect the non-linear regime of structure formation while leaving the linear regime mostly untouched. Depending on the amplitude of the sPNG, such models can either yield back the $\Lambda$CDM expectation for the power spectrum in the non-linear regime, and hence an $S_8$ parameter compatible with $\Lambda$CDM, or can even reconcile the value of $S_8$ from old weak-lensing measurements with the CMB, while solving the Hubble tension in all cases. In such models, both tensions would then be entirely related to pre-CMB physics of the early Universe. | [
{
"version": "v1",
"created": "Thu, 20 Feb 2025 14:46:30 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:00:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Stahl",
"Clément",
""
],
[
"Poulin",
"Vivian",
""
],
[
"Famaey",
"Benoit",
""
],
[
"Ibata",
"Rodrigo",
""
]
] |
2502.14746 | Nolan Coble | Nolan J. Coble, Alexander Barg | Coxeter codes: Extending the Reed-Muller family | This version, v2, is a a full version of the previous submission, v1, with additional results and full proofs added | null | null | null | cs.IT math.CO math.IT quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Binary Reed-Muller (RM) codes are defined via evaluations of Boolean-valued functions on $\mathbb{Z}_2^m$. We introduce a class of binary linear codes that generalizes the RM family by replacing the domain $\mathbb{Z}_2^m$ with an arbitrary finite Coxeter group. Like RM codes, this class is closed under duality, forms a nested code sequence, satisfies a multiplication property, and has asymptotic rate determined by a Gaussian distribution. Coxeter codes also give rise to a family of quantum codes for which transversal diagonal $Z$ rotations can perform non-trivial logic. | [
{
"version": "v1",
"created": "Thu, 20 Feb 2025 17:16:28 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 03:40:56 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Coble",
"Nolan J.",
""
],
[
"Barg",
"Alexander",
""
]
] |
2502.15533 | Alexander Jones MSc | Alexander R. Jones, Xingrui Cheng, Shravan Kumar Parthasarathy, Roland Nagy, Patrick Salter, Jason Smith, Cristian Bonato, and Christiaan Bekker | Scalable registration of single quantum emitters within solid immersion lenses through femtosecond laser writing | 7 pages, 4 figures | null | 10.1021/acs.nanolett.5c01325 | null | quant-ph physics.optics | http://creativecommons.org/licenses/by/4.0/ | Optically active silicon-vacancy (V$_{Si}$) centers in silicon carbide (SiC) serve as qubits, interfacing spins via photons. This capability allows the encoding of photonic information within the spin state and facilitates on-demand readout, promising applications such as quantum memories. However, electron irradiation, a common technique for creating defects in SiC, lacks spatial selectivity, limiting scalability. We employed femtosecond laser writing within photonic structures to generate single (V$_{Si}$) centers, registering them to photonic structures and enhancing optical collection efficiency by a factor of 4.5. Characterization of 28 laser-written defects centers in solid immersion lenses (SILs) showed distributions relative to the photonic structure's center of 260 nm in the x-direction and 60 nm in the y-direction, with standard deviations of $\pm 170$ nm and $\pm 90$ nm, respectively. This method is scalable for developing integrated quantum devices using spin-photon interfaces in SiC. | [
{
"version": "v1",
"created": "Fri, 21 Feb 2025 15:33:26 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jones",
"Alexander R.",
""
],
[
"Cheng",
"Xingrui",
""
],
[
"Parthasarathy",
"Shravan Kumar",
""
],
[
"Nagy",
"Roland",
""
],
[
"Salter",
"Patrick",
""
],
[
"Smith",
"Jason",
""
],
[
"Bonato",
"Cristian",
""
],
[
"Bekker",
"Christiaan",
""
]
] |
2502.15766 | Florian Neukart | Florian Neukart, Eike Marx, Valerii Vinokur | Reversible Imprinting and Retrieval of Quantum Information: Experimental Verification of the Quantum Memory Matrix Hypothesis | 11 pages, 2 figures | null | null | null | physics.gen-ph | http://creativecommons.org/licenses/by/4.0/ | We report the first end-to-end hardware-validated demonstration of a reversible Quantum Memory Matrix QMM imprint retrieval cycle. Using IBM Quantum back ends, we realize five imprint retrieval experiments that scale from a minimal three-qubit cell to a five-qubit dual cycle. For every circuit, we provide Wilson score 95 percent confidence intervals, Pearson correlations, and mutual information between field and output qubits, establishing unitary reversibility well beyond statistical noise for example, r Q0 Q2 equals 0.64 plus minus 0.04, p less than 10 to the power of minus 6 in the five qubit run. Taken together, the data constitute the most stringent experimental support to date for the QMM hypothesis: finite dimensional Planck scale cells can faithfully store, propagate, and return quantum information. Our results strengthen the standing of QMM as a viable, local, and unitary framework for addressing fundamental questions such as the black hole information paradox. | [
{
"version": "v1",
"created": "Sat, 15 Feb 2025 00:29:33 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 00:00:08 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Neukart",
"Florian",
""
],
[
"Marx",
"Eike",
""
],
[
"Vinokur",
"Valerii",
""
]
] |
2502.16033 | Qianqi Yan | Qianqi Yan, Yue Fan, Hongquan Li, Shan Jiang, Yang Zhao, Xinze Guan, Ching-Chen Kuo, Xin Eric Wang | Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by/4.0/ | Existing Multimodal Large Language Models (MLLMs) are predominantly trained and tested on consistent visual-textual inputs, leaving open the question of whether they can handle inconsistencies in real-world, layout-rich content. To bridge this gap, we propose the Multimodal Inconsistency Reasoning (MMIR) benchmark to assess MLLMs' ability to detect and reason about semantic mismatches in artifacts such as webpages, presentation slides, and posters. MMIR comprises 534 challenging samples, each containing synthetically injected errors across five reasoning-heavy categories: Factual Contradiction, Identity Misattribution, Contextual Mismatch, Quantitative Discrepancy, and Temporal/Spatial Incoherence. We evaluate six state-of-the-art MLLMs, showing that models with dedicated multimodal reasoning capabilities, such as o1, substantially outperform their counterparts while open-source models remain particularly vulnerable to inconsistency errors. Detailed error analyses further show that models excel in detecting pairwise inconsistencies but struggle with inconsistencies confined to single elements in complex layouts. Probing experiments reveal that single-modality prompting, including Chain-of-Thought (CoT) and Set-of-Mark (SoM) methods, yields marginal gains, revealing a key bottleneck in cross-modal reasoning. Our findings highlight the need for advanced multimodal reasoning and point to future research on multimodal inconsistency. | [
{
"version": "v1",
"created": "Sat, 22 Feb 2025 01:52:37 GMT"
},
{
"version": "v2",
"created": "Tue, 04 Mar 2025 08:23:58 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 18:32:17 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Yan",
"Qianqi",
""
],
[
"Fan",
"Yue",
""
],
[
"Li",
"Hongquan",
""
],
[
"Jiang",
"Shan",
""
],
[
"Zhao",
"Yang",
""
],
[
"Guan",
"Xinze",
""
],
[
"Kuo",
"Ching-Chen",
""
],
[
"Wang",
"Xin Eric",
""
]
] |
2502.16670 | Jukka Ruohonen | Jukka Ruohonen and Qusai Ramadan | The Popularity Hypothesis in Software Security: A Large-Scale Replication with PHP Packages | Resubmitted | null | null | null | cs.SE cs.CR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | There has been a long-standing hypothesis that a software's popularity is related to its security or insecurity in both research and popular discourse. There are also a few empirical studies that have examined the hypothesis, either explicitly or implicitly. The present work continues with and contributes to this research with a replication-motivated large-scale analysis of software written in the PHP programming language. The dataset examined contains nearly four hundred thousand open source software packages written in PHP. According to the results based on reported security vulnerabilities, the hypothesis does holds; packages having been affected by vulnerabilities over their release histories are generally more popular than packages without having been affected by a single vulnerability. With this replication results, the paper contributes to the efforts to strengthen the empirical knowledge base in cyber and software security. | [
{
"version": "v1",
"created": "Sun, 23 Feb 2025 18:00:57 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 16:43:17 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Ruohonen",
"Jukka",
""
],
[
"Ramadan",
"Qusai",
""
]
] |
2502.16794 | Xilin Jiang | Xilin Jiang, Sukru Samet Dindar, Vishal Choudhari, Stephan Bickel, Ashesh Mehta, Guy M McKhann, Daniel Friedman, Adeen Flinker, Nima Mesgarani | AAD-LLM: Neural Attention-Driven Auditory Scene Understanding | Accepted by ACL 2025 Main Conference | null | null | null | cs.SD cs.AI cs.CL cs.HC eess.AS | http://creativecommons.org/licenses/by/4.0/ | Auditory foundation models, including auditory large language models (LLMs), process all sound inputs equally, independent of listener perception. However, human auditory perception is inherently selective: listeners focus on specific speakers while ignoring others in complex auditory scenes. Existing models do not incorporate this selectivity, limiting their ability to generate perception-aligned responses. To address this, we introduce Intention-Informed Auditory Scene Understanding (II-ASU) and present Auditory Attention-Driven LLM (AAD-LLM), a prototype system that integrates brain signals to infer listener attention. AAD-LLM extends an auditory LLM by incorporating intracranial electroencephalography (iEEG) recordings to decode which speaker a listener is attending to and refine responses accordingly. The model first predicts the attended speaker from neural activity, then conditions response generation on this inferred attentional state. We evaluate AAD-LLM on speaker description, speech transcription and extraction, and question answering in multitalker scenarios, with both objective and subjective ratings showing improved alignment with listener intention. By taking a first step toward intention-aware auditory AI, this work explores a new paradigm where listener perception informs machine listening, paving the way for future listener-centered auditory systems. Demo and code available: https://aad-llm.github.io. | [
{
"version": "v1",
"created": "Mon, 24 Feb 2025 03:06:45 GMT"
},
{
"version": "v2",
"created": "Fri, 14 Mar 2025 20:46:33 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 01:15:07 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jiang",
"Xilin",
""
],
[
"Dindar",
"Sukru Samet",
""
],
[
"Choudhari",
"Vishal",
""
],
[
"Bickel",
"Stephan",
""
],
[
"Mehta",
"Ashesh",
""
],
[
"McKhann",
"Guy M",
""
],
[
"Friedman",
"Daniel",
""
],
[
"Flinker",
"Adeen",
""
],
[
"Mesgarani",
"Nima",
""
]
] |
2502.16849 | Taj Jones-McCormick | Taj Jones-McCormick, Aukosh Jagannath, Subhabrata Sen | Provable Benefits of Unsupervised Pre-training and Transfer Learning via Single-Index Models | null | null | null | null | stat.ML cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Unsupervised pre-training and transfer learning are commonly used techniques to initialize training algorithms for neural networks, particularly in settings with limited labeled data. In this paper, we study the effects of unsupervised pre-training and transfer learning on the sample complexity of high-dimensional supervised learning. Specifically, we consider the problem of training a single-layer neural network via online stochastic gradient descent. We establish that pre-training and transfer learning (under concept shift) reduce sample complexity by polynomial factors (in the dimension) under very general assumptions. We also uncover some surprising settings where pre-training grants exponential improvement over random initialization in terms of sample complexity. | [
{
"version": "v1",
"created": "Mon, 24 Feb 2025 05:13:11 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 17:36:14 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jones-McCormick",
"Taj",
""
],
[
"Jagannath",
"Aukosh",
""
],
[
"Sen",
"Subhabrata",
""
]
] |
2502.16969 | Edmund Christian Herenz | E.C. Herenz, H. Kusakabe, S. Maulick | The extreme starburst J1044+0353 blows kpc-scale bubbles | Accepted for publication in PASJ after moderate revision. (6 pages, 6 figures) | null | null | null | astro-ph.GA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We report the discovery of a kpc-scale lacy filamentary structure in low surface-brightness H$\alpha$ and [OIII]$\lambda5007$ emission around the low-redshift, extremely metal-poor and compact reionisation-era analogue SDSS J1044+0353. We identify seven elliptical arcs in H$\alpha$ emission at $\mathrm{SB}_\mathrm{H\alpha} \sim 1 - 2 \times 10^{-18}$erg s$^{-1}$cm$^{-2}$arcsec$^{-2}$. We interpret these features as limb-brightened giant-shells that bound egg-shaped super bubbles with 3-3.5 kpc radii. These shells are significantly larger than the known giant shells around nearby star-forming dwarfs. Kinematic maps reveal a gradient perpendicular to the major-axis and line broadening in the outskirts. The latter, when interpreted due to line-splitting of the expanding shells, suggests expansion velocities of $\sim 40$km s$^{-1}$. Notably, the properties of the giant shells evade description by simple analytic prescriptions for energy- and momentum-driven super-bubbles. | [
{
"version": "v1",
"created": "Mon, 24 Feb 2025 08:49:21 GMT"
},
{
"version": "v2",
"created": "Thu, 06 Mar 2025 11:02:37 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 15:13:04 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Herenz",
"E. C.",
""
],
[
"Kusakabe",
"H.",
""
],
[
"Maulick",
"S.",
""
]
] |
2502.17361 | Siyang Liu | Han-Jia Ye and Si-Yang Liu and Wei-Lun Chao | A Closer Look at TabPFN v2: Understanding Its Strengths and Extending Its Capabilities | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Tabular datasets are inherently heterogeneous, presenting significant challenges for developing pre-trained foundation models. The recently introduced transformer-based Tabular Prior-data Fitted Network v2 (TabPFN v2) achieves unprecedented in-context learning performance across diverse downstream datasets, marking a pivotal advancement in tabular foundation models. In this paper, we take a closer look at TabPFN v2 to examine how it effectively handles heterogeneity and achieves high predictive accuracy, and to explore how its limitations in high-dimensional, many-category, and large-scale tasks can be mitigated. We find that TabPFN v2 can infer attribute relationships even when provided with randomized attribute token inputs, eliminating the need to explicitly learn dataset-specific attribute embeddings to address heterogeneity. We further show that TabPFN v2 can be transformed into a feature extractor, revealing its ability to construct a highly separable feature space for accurate predictions. Lastly, we demonstrate that TabPFN v2's limitations can be addressed through a test-time divide-and-conquer strategy, enabling scalable inference without requiring re-training. By uncovering the mechanisms behind TabPFN v2's success and introducing strategies to extend its applicability, this study offers key insights into the design of future tabular foundation models. | [
{
"version": "v1",
"created": "Mon, 24 Feb 2025 17:38:42 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 04:51:44 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Ye",
"Han-Jia",
""
],
[
"Liu",
"Si-Yang",
""
],
[
"Chao",
"Wei-Lun",
""
]
] |
2502.18377 | Adeel Pervez | Adeel Pervez, Efstratios Gavves, Francesco Locatello | Mechanistic PDE Networks for Discovery of Governing Equations | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | We present Mechanistic PDE Networks -- a model for discovery of governing partial differential equations from data. Mechanistic PDE Networks represent spatiotemporal data as space-time dependent linear partial differential equations in neural network hidden representations. The represented PDEs are then solved and decoded for specific tasks. The learned PDE representations naturally express the spatiotemporal dynamics in data in neural network hidden space, enabling increased power for dynamical modeling. Solving the PDE representations in a compute and memory-efficient way, however, is a significant challenge. We develop a native, GPU-capable, parallel, sparse, and differentiable multigrid solver specialized for linear partial differential equations that acts as a module in Mechanistic PDE Networks. Leveraging the PDE solver, we propose a discovery architecture that can discover nonlinear PDEs in complex settings while also being robust to noise. We validate PDE discovery on a number of PDEs, including reaction-diffusion and Navier-Stokes equations. | [
{
"version": "v1",
"created": "Tue, 25 Feb 2025 17:21:44 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 10:04:07 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Pervez",
"Adeel",
""
],
[
"Gavves",
"Efstratios",
""
],
[
"Locatello",
"Francesco",
""
]
] |
2502.18470 | Dazhou Yu | Dazhou Yu, Riyang Bao, Ruiyu Ning, Jinghong Peng, Gengchen Mai, Liang Zhao | Spatial-RAG: Spatial Retrieval Augmented Generation for Real-World Geospatial Reasoning Questions | null | null | null | null | cs.IR cs.ET cs.LG | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Answering real-world geospatial questions--such as finding restaurants along a travel route or amenities near a landmark--requires reasoning over both geographic relationships and semantic user intent. However, existing large language models (LLMs) lack spatial computing capabilities and access to up-to-date, ubiquitous real-world geospatial data, while traditional geospatial systems fall short in interpreting natural language. To bridge this gap, we introduce Spatial-RAG, a Retrieval-Augmented Generation (RAG) framework designed for geospatial question answering. Spatial-RAG integrates structured spatial databases with LLMs via a hybrid spatial retriever that combines sparse spatial filtering and dense semantic matching. It formulates the answering process as a multi-objective optimization over spatial and semantic relevance, identifying Pareto-optimal candidates and dynamically selecting the best response based on user intent. Experiments across multiple tourism and map-based QA datasets show that Spatial-RAG significantly improves accuracy, precision, and ranking performance over strong baselines. | [
{
"version": "v1",
"created": "Tue, 04 Feb 2025 01:30:06 GMT"
},
{
"version": "v2",
"created": "Thu, 27 Feb 2025 05:17:57 GMT"
},
{
"version": "v3",
"created": "Fri, 14 Mar 2025 02:48:55 GMT"
},
{
"version": "v4",
"created": "Thu, 05 Jun 2025 05:50:15 GMT"
},
{
"version": "v5",
"created": "Wed, 11 Jun 2025 04:41:29 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Yu",
"Dazhou",
""
],
[
"Bao",
"Riyang",
""
],
[
"Ning",
"Ruiyu",
""
],
[
"Peng",
"Jinghong",
""
],
[
"Mai",
"Gengchen",
""
],
[
"Zhao",
"Liang",
""
]
] |
2502.18917 | Chuyue Sun | Chuyue Sun, Viraj Agashe, Saikat Chakraborty, Jubi Taneja, Clark Barrett, David Dill, Xiaokang Qiu, Shuvendu K. Lahiri | ClassInvGen: Class Invariant Synthesis using Large Language Models | null | null | null | null | cs.PL cs.SE | http://creativecommons.org/licenses/by/4.0/ | Formal program specifications in the form of preconditions, postconditions, and class invariants have several benefits for the construction and maintenance of programs. They not only aid in program understanding due to their unambiguous semantics but can also be enforced dynamically (or even statically when the language supports a formal verifier). However, synthesizing high-quality specifications in an underlying programming language is limited by the expressivity of the specifications or the need to express them in a declarative manner. Prior work has demonstrated the potential of large language models (LLMs) for synthesizing high-quality method pre/postconditions for Python and Java, but does not consider class invariants.
In this work, we describe ClassInvGen, a method for co-generating executable class invariants and test inputs to produce high-quality class invariants for a mainstream language such as C++, leveraging LLMs' ability to synthesize pure functions. We show that ClassInvGen outperforms a pure LLM-based technique to generate specifications (from code) as well as prior data-driven invariant inference techniques such as Daikon. We contribute a benchmark of standard C++ data structures along with a harness that can help measure both the correctness and completeness of generated specifications using tests and mutants. We also demonstrate its applicability to real-world code by performing a case study on several classes within a widely used and high-integrity C++ codebase. | [
{
"version": "v1",
"created": "Wed, 26 Feb 2025 08:10:57 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 22:11:53 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Sun",
"Chuyue",
""
],
[
"Agashe",
"Viraj",
""
],
[
"Chakraborty",
"Saikat",
""
],
[
"Taneja",
"Jubi",
""
],
[
"Barrett",
"Clark",
""
],
[
"Dill",
"David",
""
],
[
"Qiu",
"Xiaokang",
""
],
[
"Lahiri",
"Shuvendu K.",
""
]
] |
2502.19324 | Simone Rossoni | Simone Rossoni and G\"unter Sigl | Anisotropy signal of UHECRs from a structured magnetized Universe | 34 pages and 20 figures | null | null | null | astro-ph.HE | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The surprising isotropy of the ultra-high-energy cosmic ray (UHECR) sky makes it difficult to identify their sources. Observables such as energy spectrum, mass composition and arrival directions are affected by interactions with background photon fields and by deflection in the extragalactic and galactic magnetic fields (EGMF and GMF). In this work, we simulate the propagation of UHECRs with energy above 8 EeV in magnetized replicas of the local Universe, obtained from constrained simulations of the Large Scale Structure. We obtain the real magnetic deflection in structured EGMF models with realistic three-dimensional simulations. We investigate different scenarios for the UHECR source distributions and densities. The effect of the GMF can be different depending on the field model considered. In this work we consider the JF12 model by mapping the arrival directions at the edge of the galaxy to those at Earth. We study the arrival direction distribution of the propagated UHECRs, and in particular their angular power spectrum, dipole and quadrupole moments. We find that the properties of the source distribution affect the cosmic ray anisotropy more than the EGMF model considered. In particular, the low multipole components depend on both the source distribution and the density. We also find that it is difficult to simultaneously reproduce the observed dipole and quadrupole values above EeV. In general, we predict too large a quadrupole strength, incompatible with observations. | [
{
"version": "v1",
"created": "Wed, 26 Feb 2025 17:16:28 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:30:37 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Rossoni",
"Simone",
""
],
[
"Sigl",
"Günter",
""
]
] |
2502.19409 | Ingo Ziegler | Danae S\'anchez Villegas, Ingo Ziegler, Desmond Elliott | ImageChain: Advancing Sequential Image-to-Text Reasoning in Multimodal Large Language Models | Code, dataset, and checkpoints are publicly available at https://github.com/danaesavi/ImageChain; v2: added human annotation study to validate SimRate | null | null | null | cs.CV cs.CL cs.LG | http://creativecommons.org/licenses/by/4.0/ | Reasoning over sequences of images remains a challenge for multimodal large language models (MLLMs). While recent models incorporate multi-image data during pre-training, they still struggle to recognize sequential structures, often treating images independently. This work introduces ImageChain, a framework that enhances MLLMs with sequential reasoning capabilities over image data by modeling visual sequences as a multi-turn conversation. In ImageChain, images are interleaved with corresponding textual descriptions to form a controlled dialogue that explicitly captures temporal dependencies and narrative progression. Our method optimizes for the task of next-scene description, where the model generates a context-aware description of an upcoming scene based on preceding visual and textual cues. We demonstrate that our approach improves performance on the next-scene description task -- achieving an average improvement from 3.7% to 19% in SimRate, a metric that quantifies semantic similarity to human-annotated ground truths. Moreover, ImageChain achieves robust zero-shot out-of-domain performance in applications ranging from comics to robotics. Extensive experiments validate that instruction-tuning in a multimodal, multi-turn conversation design is key to bridging the gap between static image understanding and temporally-aware reasoning. | [
{
"version": "v1",
"created": "Wed, 26 Feb 2025 18:55:06 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:35:11 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Villegas",
"Danae Sánchez",
""
],
[
"Ziegler",
"Ingo",
""
],
[
"Elliott",
"Desmond",
""
]
] |
2502.19826 | Shangfei Wu | Shangfei Wu, Xianghan Xu, Fei-Ting Huang, Turan Birol, Sang-Wook Cheong, Girsh Blumberg | Phonon anomalies within the polar charge density wave phase of the structurally chiral superconductor Mo$_3$Al$_2$C | 10 pages, 6 figures. The supplemental materials are included. Journal version | Physical Review B 111, 224505 (2025) | 10.1103/nkx2-6qyr | null | cond-mat.supr-con cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We employ polarization-resolved Raman spectroscopy to study the lattice dynamics of the polar charge density wave phase of the superconductor Mo$_3$Al$_2$C with structural chirality. We show the phononic signatures of the charge density wave transition at $T^*$ = 155\,K in Mo$_3$Al$_2$C. The detailed temperature dependence of these phonon modes' frequency, half width at half maximum, and integrated area below $T^*$ reveal anomalies at an intermediate temperature $T' \sim$ 100\,K, especially for the low-energy modes at 130 and 180\,\cm-1. We discuss the origin of these phonon anomalies within the polar charge density wave phase of Mo$_3$Al$_2$C. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 07:00:42 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:52:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Wu",
"Shangfei",
""
],
[
"Xu",
"Xianghan",
""
],
[
"Huang",
"Fei-Ting",
""
],
[
"Birol",
"Turan",
""
],
[
"Cheong",
"Sang-Wook",
""
],
[
"Blumberg",
"Girsh",
""
]
] |
2502.19830 | Yiwei Li | Yiwei Li, Ji Zhang, Shaoxiong Feng, Peiwen Yuan, Xinglin Wang, Jiayi Shi, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li | Revisiting Self-Consistency from Dynamic Distributional Alignment Perspective on Answer Aggregation | ACL 2025 Findings | null | null | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Self-consistency improves reasoning by aggregating diverse stochastic samples, yet the dynamics behind its efficacy remain underexplored. We reframe self-consistency as a dynamic distributional alignment problem, revealing that decoding temperature not only governs sampling randomness but also actively shapes the latent answer distribution. Given that high temperatures require prohibitively large sample sizes to stabilize, while low temperatures risk amplifying biases, we propose a confidence-driven mechanism that dynamically calibrates temperature: sharpening the sampling distribution under uncertainty to align with high-probability modes, and promoting exploration when confidence is high. Experiments on mathematical reasoning tasks show this approach outperforms fixed-diversity baselines under limited samples, improving both average and best-case performance across varying initial temperatures without additional data or modules. This establishes self-consistency as a synchronization challenge between sampling dynamics and evolving answer distributions. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 07:07:40 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 07:11:03 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Li",
"Yiwei",
""
],
[
"Zhang",
"Ji",
""
],
[
"Feng",
"Shaoxiong",
""
],
[
"Yuan",
"Peiwen",
""
],
[
"Wang",
"Xinglin",
""
],
[
"Shi",
"Jiayi",
""
],
[
"Zhang",
"Yueqi",
""
],
[
"Tan",
"Chuyi",
""
],
[
"Pan",
"Boyuan",
""
],
[
"Hu",
"Yao",
""
],
[
"Li",
"Kan",
""
]
] |
2502.19845 | Chengyuan Huang | Chengyuan Huang, Zhenlan Chen, Mengke Ha, Haoyuan Wang, Qing Xiao, Changjian Ma, Danqing Liu, Zhiyuan Qin, Dawei Qiu, Ziliang Guo, Dingbang Chen, Qianyi Zhao, Yanling Liu, Chengxuan Ye, Zhenhao Li, and Guanglei Cheng | A Milli-Kelvin Atomic Force Microscope Made of Glass | The following article has been submitted to Review of Scientific Instruments. After it is published, it will be found at https://pubs.aip.org/aip/rsi | Rev. Sci. Instrum. 96, 063702 (2025) | 10.1063/5.0260775 | null | physics.ins-det | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Milli-Kelvin atomic force microscopy (mK-AFM) presents an ongoing experimental challenge due to the intense vibrations in a cryogen-free dilution refrigerator and the low cooling power available at mK temperatures. A viable approach is to make the system exceptionally rigid and thermally insulating to decouple external vibrations and isolate heat dissipation from the piezo elements. Here, we present a low-cost and large scan-range mK-AFM that operates below 100 mK. All the essential parts of our mK-AFM, including the scanners, tip assembly, and microscope body, are custom-made of fused silica glass by taking advantage of its high specific modulus, extremely low thermal expansion coefficient, and excellent thermal insulation properties. We carefully balance the scan range (25 ${\mu}$m $\times$ 25 ${\mu}$m), heat dissipation, and stiffness of the system to reach optimal performance at mK temperatures. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 07:44:44 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Huang",
"Chengyuan",
""
],
[
"Chen",
"Zhenlan",
""
],
[
"Ha",
"Mengke",
""
],
[
"Wang",
"Haoyuan",
""
],
[
"Xiao",
"Qing",
""
],
[
"Ma",
"Changjian",
""
],
[
"Liu",
"Danqing",
""
],
[
"Qin",
"Zhiyuan",
""
],
[
"Qiu",
"Dawei",
""
],
[
"Guo",
"Ziliang",
""
],
[
"Chen",
"Dingbang",
""
],
[
"Zhao",
"Qianyi",
""
],
[
"Liu",
"Yanling",
""
],
[
"Ye",
"Chengxuan",
""
],
[
"Li",
"Zhenhao",
""
],
[
"Cheng",
"Guanglei",
""
]
] |
2502.20181 | Wenkang Jiang | Wenkang Jiang, Jiaxin Han, Fuyu Dong, Feihong He | Self-similar decomposition of the hierarchical merger tree of dark matter halos | 20 pages, 13 figures | null | null | null | astro-ph.CO astro-ph.GA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In the $\Lambda$CDM universe, structure formation is generally not a self-similar process, while some self-similarity remains in certain statistics which can greatly simplify our description and understanding of the cosmic structures. In this work, we show that the merger tree of dark matter halos is approximately self-similar by investigating the universality of the subhalo peak mass function (PMF) describing the mass distribution of progenitor halos. Using a set of cosmological simulations and identifying subhalos of different merger levels with HBT+, we verify that the level-1 subhalo PMF is close to universal across halo mass, redshift, and cosmology. This approximate self-similarity allows us to analytically derive the subhalo PMF for subhalos accreted at any level (i.e., for sub-sub...halos) through self-convolutions of the level-1 PMF, and the resulting model shows good agreement with simulation measurements. We further derive a number of analytical properties on the hierarchical origin of subhalos. We show that higher-level subhalos dominate at progressively lower peak mass in the PMF and are more likely to originate from major mergers than lower-level ones. At a given merger mass ratio, the subhalo accretion rates at each level track the growth rate of the host halo. At a fixed final mass ratio, however, subhalos of higher-level, higher-mass-ratio and in more massive haloes tend to be accreted more recently. Matching subhalo peak mass to galaxy mass, these results have direct implications on the hierarchical origin of satellite galaxies. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 15:18:28 GMT"
},
{
"version": "v2",
"created": "Fri, 28 Feb 2025 07:30:21 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 13:56:00 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jiang",
"Wenkang",
""
],
[
"Han",
"Jiaxin",
""
],
[
"Dong",
"Fuyu",
""
],
[
"He",
"Feihong",
""
]
] |
2502.20266 | Kohei Hattori | Kohei Hattori, Daisuke Taniguchi, Takuji Tsujimoto, Noriyuki Matsunaga, Hiroaki Sameshima, Scarlet S. Elgueta, Shogo Otsubo | Discovery of WINERED-HVS1: A metal-rich hyper-velocity star candidate ejected from the Galactic center | 37 pages (main text 22 pages), 16 figures, 6 tables. ApJ accepted | null | null | null | astro-ph.GA astro-ph.SR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We report the discovery of a metal-rich red giant star, WINERED-HVS1, which is a candidate for a hyper-velocity star (HVS). Its past trajectory suggests that this star may have been ejected by the Galactic supermassive black hole (SMBH; Sgr A*), with a modest ejection velocity of at least $500\; \mathrm{km\;s^{-1}}$. Since WINERED-HVS1 is gravitationally bound to the Milky Way and is not necessarily young, it is not an unambiguous HVS candidate from a kinematic perspective, unlike the previously confirmed A-type main-sequence HVS known as S5-HVS1. Therefore, the chemical characterization of this star is essential for understanding its origin. Through a spectroscopic follow-up observation with Magellan/WINERED, we determined its metallicity $\mathrm{[Fe/H]}=-0.147^{+0.046}_{-0.045}$, as well as the abundances for Na, Mg, Si, S, K, Ca, Ti, Cr, Ni, and Sr. The high metallicity value suggests that this star was ejected from the Galactic center and is unlikely to be a halo star with a radial orbit. The abundance pattern of this star -- including the pattern of [$\alpha$/Fe], [Mg/Fe], [Si/Mg], and [Ca/Mg] -- is consistent with that of the nuclear star cluster surrounding the SMBH, further supporting our view. This discovery opens a new window to look through the Galactic center environment without the hindrance of dust extinction, by using the HVSs moving in the halo region where dust extinction is minimal. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 16:53:56 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 04:51:42 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Hattori",
"Kohei",
""
],
[
"Taniguchi",
"Daisuke",
""
],
[
"Tsujimoto",
"Takuji",
""
],
[
"Matsunaga",
"Noriyuki",
""
],
[
"Sameshima",
"Hiroaki",
""
],
[
"Elgueta",
"Scarlet S.",
""
],
[
"Otsubo",
"Shogo",
""
]
] |
2502.20383 | Seungjae Lee | Jeffrey Yang Fan Chiang, Seungjae Lee, Jia-Bin Huang, Furong Huang, Yizheng Chen | Why Are Web AI Agents More Vulnerable Than Standalone LLMs? A Security Analysis | Project website: http://vulnerable-ai-agents.github.io | null | null | null | cs.LG cs.CL | http://creativecommons.org/licenses/by/4.0/ | Recent advancements in Web AI agents have demonstrated remarkable capabilities in addressing complex web navigation tasks. However, emerging research shows that these agents exhibit greater vulnerability compared to standalone Large Language Models (LLMs), despite both being built upon the same safety-aligned models. This discrepancy is particularly concerning given the greater flexibility of Web AI Agent compared to standalone LLMs, which may expose them to a wider range of adversarial user inputs. To build a scaffold that addresses these concerns, this study investigates the underlying factors that contribute to the increased vulnerability of Web AI agents. Notably, this disparity stems from the multifaceted differences between Web AI agents and standalone LLMs, as well as the complex signals - nuances that simple evaluation metrics, such as success rate, often fail to capture. To tackle these challenges, we propose a component-level analysis and a more granular, systematic evaluation framework. Through this fine-grained investigation, we identify three critical factors that amplify the vulnerability of Web AI agents; (1) embedding user goals into the system prompt, (2) multi-step action generation, and (3) observational capabilities. Our findings highlights the pressing need to enhance security and robustness in AI agent design and provide actionable insights for targeted defense strategies. | [
{
"version": "v1",
"created": "Thu, 27 Feb 2025 18:56:26 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 03:01:49 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Chiang",
"Jeffrey Yang Fan",
""
],
[
"Lee",
"Seungjae",
""
],
[
"Huang",
"Jia-Bin",
""
],
[
"Huang",
"Furong",
""
],
[
"Chen",
"Yizheng",
""
]
] |
2502.20786 | Yuhang Zhang | Yuhang Zhang and Minghui Song | Dimension-independent convergence rate of propagation of chaos and numerical analysis for McKean-Vlasov stochastic differential equations with coefficients nonlinearly dependent on measure | null | null | null | null | math.NA cs.NA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In contrast to ordinary stochastic differential equations (SDEs), the numerical simulation of McKean-Vlasov stochastic differential equations (MV-SDEs) requires approximating the distribution law first. Based on the theory of propagation of chaos, particle approximation method is widely used. Then, a natural question is to investigate the convergence rate of the method (also referred to as the convergence rate of PoC). In fact, the PoC convergence rate is well understood for MV-SDEs with coefficients linearly dependent on the measure, but the rate deteriorates with dimension $d$ under the $L^p$-Wasserstein metric for nonlinear measure-dependent coefficients, even when Lipschitz continuity with respect to the measure is assumed. The main objective of this paper is to establish a dimension-independent convergence result of PoC for MV-SDEs whose coefficients are nonlinear with respect to the measure component but Lipschitz continuous. As a complement we further give the time discretization of the equations and thus verify the convergence rate of PoC using numerical experiments. | [
{
"version": "v1",
"created": "Fri, 28 Feb 2025 07:07:23 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 10:24:32 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Zhang",
"Yuhang",
""
],
[
"Song",
"Minghui",
""
]
] |
2502.20838 | Md Ragib Amin Nihal | Ragib Amin Nihal, Benjamin Yen, Runwu Shi, Kazuhiro Nakadai | Weakly Supervised Multiple Instance Learning for Whale Call Detection and Temporal Localization in Long-Duration Passive Acoustic Monitoring | null | null | null | null | cs.SD cs.AI cs.LG eess.AS | http://creativecommons.org/licenses/by/4.0/ | Marine ecosystem monitoring via Passive Acoustic Monitoring (PAM) generates vast data, but deep learning often requires precise annotations and short segments. We introduce DSMIL-LocNet, a Multiple Instance Learning framework for whale call detection and localization using only bag-level labels. Our dual-stream model processes 2-30 minute audio segments, leveraging spectral and temporal features with attention-based instance selection. Tests on Antarctic whale data show longer contexts improve classification (F1: 0.8-0.9) while medium instances ensure localization precision (0.65-0.70). This suggests MIL can enhance scalable marine monitoring. Code: https://github.com/Ragib-Amin-Nihal/DSMIL-Loc | [
{
"version": "v1",
"created": "Fri, 28 Feb 2025 08:34:12 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 06:25:19 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Nihal",
"Ragib Amin",
""
],
[
"Yen",
"Benjamin",
""
],
[
"Shi",
"Runwu",
""
],
[
"Nakadai",
"Kazuhiro",
""
]
] |
2502.21059 | Zhen Sun | Ziyi Zhang, Zhen Sun, Zongmin Zhang, Jihui Guo, Xinlei He | FC-Attack: Jailbreaking Multimodal Large Language Models via Auto-Generated Flowcharts | 13 pages, 7 figures | null | null | null | cs.CV cs.AI cs.CR cs.LG | http://creativecommons.org/licenses/by/4.0/ | Multimodal Large Language Models (MLLMs) have become powerful and widely adopted in some practical applications. However, recent research has revealed their vulnerability to multimodal jailbreak attacks, whereby the model can be induced to generate harmful content, leading to safety risks. Although most MLLMs have undergone safety alignment, recent research shows that the visual modality is still vulnerable to jailbreak attacks. In our work, we discover that by using flowcharts with partially harmful information, MLLMs can be induced to provide additional harmful details. Based on this, we propose a jailbreak attack method based on auto-generated flowcharts, FC-Attack. Specifically, FC-Attack first fine-tunes a pre-trained LLM to create a step-description generator based on benign datasets. The generator is then used to produce step descriptions corresponding to a harmful query, which are transformed into flowcharts in 3 different shapes (vertical, horizontal, and S-shaped) as visual prompts. These flowcharts are then combined with a benign textual prompt to execute the jailbreak attack on MLLMs. Our evaluations on Advbench show that FC-Attack attains an attack success rate of up to 96% via images and up to 78% via videos across multiple MLLMs. Additionally, we investigate factors affecting the attack performance, including the number of steps and the font styles in the flowcharts. We also find that FC-Attack can improve the jailbreak performance from 4% to 28% in Claude-3.5 by changing the font style. To mitigate the attack, we explore several defenses and find that AdaShield can largely reduce the jailbreak performance but with the cost of utility drop. | [
{
"version": "v1",
"created": "Fri, 28 Feb 2025 13:59:11 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 18:09:45 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Zhang",
"Ziyi",
""
],
[
"Sun",
"Zhen",
""
],
[
"Zhang",
"Zongmin",
""
],
[
"Guo",
"Jihui",
""
],
[
"He",
"Xinlei",
""
]
] |
2502.21075 | Christopher Wewer | Christopher Wewer, Bart Pogodzinski, Bernt Schiele, Jan Eric Lenssen | Spatial Reasoning with Denoising Models | Project website: https://geometric-rl.mpi-inf.mpg.de/srm | null | null | null | cs.CV cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We introduce Spatial Reasoning Models (SRMs), a framework to perform reasoning over sets of continuous variables via denoising generative models. SRMs infer continuous representations on a set of unobserved variables, given observations on observed variables. Current generative models on spatial domains, such as diffusion and flow matching models, often collapse to hallucination in case of complex distributions. To measure this, we introduce a set of benchmark tasks that test the quality of complex reasoning in generative models and can quantify hallucination. The SRM framework allows to report key findings about importance of sequentialization in generation, the associated order, as well as the sampling strategies during training. It demonstrates, for the first time, that order of generation can successfully be predicted by the denoising network itself. Using these findings, we can increase the accuracy of specific reasoning tasks from <1% to >50%. Our project website provides additional videos, code, and the benchmark datasets: https://geometric-rl.mpi-inf.mpg.de/srm | [
{
"version": "v1",
"created": "Fri, 28 Feb 2025 14:08:30 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 18:54:54 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Wewer",
"Christopher",
""
],
[
"Pogodzinski",
"Bart",
""
],
[
"Schiele",
"Bernt",
""
],
[
"Lenssen",
"Jan Eric",
""
]
] |
2503.00469 | Klaus Bartschat | J. Jakob, C. Bauer, M.-J. Ilhan, D. Bharti, C. Ott, T. Pfeifer, K. Bartschat, A. Harth | Extracting RABBITT-like phase information from time-dependent transient absorption spectra | 7 pages, 5 figures | Phys. Rev. Res. 7 (2025) 023244 | 10.1103/PhysRevResearch.7.023244 | null | physics.atom-ph quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We explore how the spectral phase of atto\-second pulse trains influences the optical cross section in transient absorption (TA) spectroscopy. The interaction of extreme ultraviolet (XUV) and time-delayed near-infrared (NIR) fields with an atomic or molecular system governs the dynamics. As already shown in RABBITT experiments, the spectral phase of the XUV pulses can be extracted from the photoionization spectrum as a function of the time delay. Similarly, this XUV phase imprints itself on delay-dependent optical cross-section oscillations. With a perturbative analytical approach and by simulating the quantum dynamics both in a few-level model and via solving the time-dependent Schr\"odinger equation for atomic hydrogen, we reveal the similarity between the spectral phase in RABBITT and TA spectroscopy. | [
{
"version": "v1",
"created": "Sat, 01 Mar 2025 12:48:02 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Jakob",
"J.",
""
],
[
"Bauer",
"C.",
""
],
[
"Ilhan",
"M. -J.",
""
],
[
"Bharti",
"D.",
""
],
[
"Ott",
"C.",
""
],
[
"Pfeifer",
"T.",
""
],
[
"Bartschat",
"K.",
""
],
[
"Harth",
"A.",
""
]
] |
2503.01108 | Andrew Yi | Byeong Rok Ko, Andrew K. Yi | Asymmetrical cavity design that bypasses mode mixings in axion haloscope experiments | 5 pages, 5 figures | null | null | null | hep-ex hep-ph physics.ins-det | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Microwave cavities used in axion haloscope experiments typically employ a tuning rod as a means to widen the range of resonance frequencies at which it is sensitive to axion-to-photon conversion. A realistic tuning mechanism requires a gap between the cavity end caps and the tuning rod to ensure movement, and causes some modes to hybridize with the resonant mode that is being tracked for the experiment. These so-called mode mixings lead to gaps in the frequency range that practically lose sensitivity to axions. In order to solve this problem, we present a cavity design which, for two tuning rod configurations corresponding to a lower and higher frequency range, have a dielectric rod inserted at a specific location that makes the cavity asymmetrical. Moving the tuning rod closer to the dielectric insert changes the location and frequency of the mode mixing compared to when it is farther away from it. This design is easily realizable in practical experiments and makes possible an axion dark matter search with minimal loss in sensitivity due to mode mixings. We also show that the same design has the same desired effect when cavity dimensions are scaled down to be smaller and are at higher resonance frequencies. | [
{
"version": "v1",
"created": "Mon, 03 Mar 2025 02:31:24 GMT"
},
{
"version": "v2",
"created": "Thu, 22 May 2025 06:59:57 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 05:44:51 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Ko",
"Byeong Rok",
""
],
[
"Yi",
"Andrew K.",
""
]
] |
2503.01940 | Xuan Zhang | Xuan Zhang, Yongliang Shen, Zhe Zheng, Linjuan Wu, Wenqi Zhang, Yuchen Yan, Qiuying Peng, Jun Wang, Weiming Lu | AskToAct: Enhancing LLMs Tool Use via Self-Correcting Clarification | null | null | null | null | cs.CL cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large language models (LLMs) have demonstrated remarkable capabilities in tool learning. In real-world scenarios, user queries are often ambiguous and incomplete, requiring effective clarification. However, existing interactive clarification approaches face two critical limitations: reliance on manually constructed datasets, which inherently constrains training data scale and diversity, and lack of error correction mechanisms during multi-turn clarification, leading to error accumulation that compromises both accuracy and efficiency. We present AskToAct, which addresses these challenges by exploiting the structural mapping between queries and their tool invocation solutions. Our key insight is that tool parameters naturally represent explicit user intents. By systematically removing key parameters from queries while retaining them as ground truth, we enable automated construction of high-quality training data. We further enhance model robustness through error-correction pairs and selective masking, enabling dynamic error detection during clarification interactions. Comprehensive experiments demonstrate that AskToAct significantly outperforms existing approaches, achieving above 57% accuracy in recovering critical unspecified intents and enhancing clarification efficiency by an average of 10.46% while maintaining high accuracy in tool invocation. Our framework exhibits robust performance across different model architectures and successfully generalizes to entirely unseen APIs without additional training, achieving performance comparable to GPT-4o with substantially fewer computational resources. | [
{
"version": "v1",
"created": "Mon, 03 Mar 2025 12:55:49 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 12:06:37 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Zhang",
"Xuan",
""
],
[
"Shen",
"Yongliang",
""
],
[
"Zheng",
"Zhe",
""
],
[
"Wu",
"Linjuan",
""
],
[
"Zhang",
"Wenqi",
""
],
[
"Yan",
"Yuchen",
""
],
[
"Peng",
"Qiuying",
""
],
[
"Wang",
"Jun",
""
],
[
"Lu",
"Weiming",
""
]
] |
2503.02079 | Paul-Hermann Balduf | Paul-Hermann Balduf | Ladders and rainbows in Minimal Subtraction | 11 pages, 5 figures. Mathematica notebook available as ancillary file | Phys. Rev. D 111, 116009 (11 June, 2025) | 10.1103/kxln-26db | null | hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In dimensional regularization with $D=D_0-2\epsilon$, the minimal subtraction (MS) scheme is characterized by counterterms that only consist of singular terms in $\epsilon$. We develop a general method to compute the infinite sums of massless ladder or rainbow Feynman integrals in MS at $D_0$. Our method is based on relating the MS-solution to a kinematic solution at a coupling-dependent renormalization point. If the $\epsilon$-dependent Mellin transform of the kernel diagram of the insertions can be computed in closed form, we typically obtain a closed form expression for the all-order solution in MS. As examples, we consider Yukawa theory and $\phi^4$ theory in $D_0=4$, and $\phi^3$ theory in $D_0=6$. | [
{
"version": "v1",
"created": "Mon, 03 Mar 2025 21:58:23 GMT"
},
{
"version": "v2",
"created": "Mon, 12 May 2025 09:41:32 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Balduf",
"Paul-Hermann",
""
]
] |
2503.02450 | Yilun Qiu | Yilun Qiu, Xiaoyan Zhao, Yang Zhang, Yimeng Bai, Wenjie Wang, Hong Cheng, Fuli Feng, Tat-Seng Chua | Measuring What Makes You Unique: Difference-Aware User Modeling for Enhancing LLM Personalization | 2025 ACL Findings | null | null | null | cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Personalizing Large Language Models (LLMs) has become a critical step in facilitating their widespread application to enhance individual life experiences. In pursuit of personalization, distilling key preference information from an individual's historical data as instructional preference context to customize LLM generation has emerged as a promising direction. However, these methods face a fundamental limitation by overlooking the inter-user comparative analysis, which is essential for identifying the inter-user differences that truly shape preferences. To address this limitation, we propose Difference-aware Personalization Learning (DPL), a novel approach that emphasizes extracting inter-user differences to enhance LLM personalization. DPL strategically selects representative users for comparison and establishes a structured standard to extract meaningful, task-relevant differences for customizing LLM generation. Extensive experiments on real-world datasets demonstrate that DPL significantly enhances LLM personalization. We release our code at https://github.com/SnowCharmQ/DPL. | [
{
"version": "v1",
"created": "Tue, 04 Mar 2025 09:53:26 GMT"
},
{
"version": "v2",
"created": "Sun, 01 Jun 2025 11:09:47 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 04:01:56 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Qiu",
"Yilun",
""
],
[
"Zhao",
"Xiaoyan",
""
],
[
"Zhang",
"Yang",
""
],
[
"Bai",
"Yimeng",
""
],
[
"Wang",
"Wenjie",
""
],
[
"Cheng",
"Hong",
""
],
[
"Feng",
"Fuli",
""
],
[
"Chua",
"Tat-Seng",
""
]
] |
2503.02493 | Savitri Gallego | Savitri Gallego, Uwe Oberlack, Jan Lommler, Christopher M. Karwin, Andreas Zoglauer, Pierre Jean, Peter von Ballmoos, Carolyn Kierans, Clio Sleator, John A. Tomsick and Steven E. Boggs | Bottom-up Background Simulations of the 2016 COSI Balloon Flight | accepted to ApJ the 9 May 2025 | null | 10.3847/1538-4357/add6a0 | null | astro-ph.IM | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The Compton Spectrometer and Imager (COSI) is a Compton telescope designed to survey the 0.2-5 MeV sky, consisting of a compact array of cross-strip germanium detectors. As part of its development, in 2016 COSI had a successful 46 day flight on board NASA's Super Pressure Balloon platform. This was a precursor to the COSI Small Explorer (COSI-SMEX) satellite mission that will launch in 2027 into a equatorial low Earth (530 km) orbit. The observation of MeV gamma-rays is dominated by background radiation, especially due to the activation of the detector materials induced by cosmic-ray interactions. Thus, background simulation and identification are crucial for the data analysis. Because the COSI-SMEX detectors will be similar to the ones used for the balloon flight, the balloon measurements provide an important tool for testing and cross-checking our background simulations for the upcoming space mission. In this work we perform Monte Carlo simulations of the background emission from the 2016 COSI balloon flight. Including a phenomenological shape correction, we obtain an agreement with the data at the 10-20% level for energies between 0.1-1.6 MeV, and we successfully reproduce most of the activation lines induced by cosmic ray interactions. | [
{
"version": "v1",
"created": "Tue, 04 Mar 2025 10:58:27 GMT"
},
{
"version": "v2",
"created": "Fri, 09 May 2025 08:45:57 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Gallego",
"Savitri",
""
],
[
"Oberlack",
"Uwe",
""
],
[
"Lommler",
"Jan",
""
],
[
"Karwin",
"Christopher M.",
""
],
[
"Zoglauer",
"Andreas",
""
],
[
"Jean",
"Pierre",
""
],
[
"von Ballmoos",
"Peter",
""
],
[
"Kierans",
"Carolyn",
""
],
[
"Sleator",
"Clio",
""
],
[
"Tomsick",
"John A.",
""
],
[
"Boggs",
"Steven E.",
""
]
] |
2503.03047 | Byron Chin | Byron Chin, Elchanan Mossel, Youngtak Sohn, Alexander S. Wein | Stochastic block models with many communities and the Kesten--Stigum bound | 46 pages, 1 figure, added discussion and minor corrections, extended abstract in COLT 2025 | null | null | null | math.PR cs.SI math.ST stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study the inference of communities in stochastic block models with a growing number of communities. For block models with $n$ vertices and a fixed number of communities $q$, it was predicted in Decelle et al. (2011) that there are computationally efficient algorithms for recovering the communities above the Kesten--Stigum (KS) bound and that efficient recovery is impossible below the KS bound. This conjecture has since stimulated a lot of interest, with the achievability side proven in a line of research that culminated in the work of Abbe and Sandon (2018). Conversely, recent work by Sohn and Wein (2025) provides evidence for the hardness part using the low-degree paradigm.
In this paper we investigate community recovery in the regime $q=q_n \to \infty$ as $n\to\infty$ where no such predictions exist. We show that efficient inference of communities remains possible above the KS bound. Furthermore, we show that recovery of block models is low-degree hard below the KS bound when the number of communities satisfies $q\ll \sqrt{n}$. Perhaps surprisingly, we find that when $q \gg \sqrt{n}$, there is an efficient algorithm based on non-backtracking walks for recovery even below the KS bound. We identify a new threshold and ask if it is the threshold for efficient recovery in this regime. Finally, we show that detection is easy and identify (up to a constant) the information-theoretic threshold for community recovery as the number of communities $q$ diverges.
Our low-degree hardness results also naturally have consequences for graphon estimation, improving results of Luo and Gao (2024). | [
{
"version": "v1",
"created": "Tue, 04 Mar 2025 23:08:51 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 17:18:47 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Chin",
"Byron",
""
],
[
"Mossel",
"Elchanan",
""
],
[
"Sohn",
"Youngtak",
""
],
[
"Wein",
"Alexander S.",
""
]
] |
2503.03243 | Ting Lin | Kaibo Hu, Ting Lin | Finite element form-valued forms: Construction | 93 pages, 24 figures, 21 tables | null | null | null | math.NA cs.NA math.DG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We provide a finite element discretization of $\ell$-form-valued $k$-forms on triangulation in $\mathbb{R}^{n}$ for general $k$, $\ell$ and $n$ and any polynomial degree. The construction generalizes finite element Whitney forms for the de~Rham complex and their higher-order and distributional versions, the Regge finite elements and the Christiansen--Regge elasticity complex, the TDNNS element for symmetric stress tensors, the MCS element for traceless matrix fields, the Hellan--Herrmann--Johnson (HHJ) elements for biharmonic equations, and discrete divdiv and Hessian complexes in [Hu, Lin, and Zhang, 2025]. The construction discretizes the Bernstein--Gelfand--Gelfand (BGG) diagrams. Applications of the construction include discretization of strain and stress tensors in continuum mechanics and metric and curvature tensors in differential geometry in any dimension. | [
{
"version": "v1",
"created": "Wed, 05 Mar 2025 07:51:15 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 01:42:35 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Hu",
"Kaibo",
""
],
[
"Lin",
"Ting",
""
]
] |
2503.03288 | Tapomoy Guha Sarkar | Pankaj Chavan, Tapomoy Guha Sarkar, Chandrachud B.V. Dash, Anjan A Sen | A semi-cosmographic approach to study cosmological evolution in phase space | Accepted for Publication in JCAP | null | null | null | astro-ph.CO | http://creativecommons.org/publicdomain/zero/1.0/ | The signature of Baryon Acoustic Oscillation in the clustering of dark-matter tracers allows us to measure $(D_A(z), H(z))$ independently. Treating these as conjugate variables, we are motivated to study cosmological evolution in the phase space of dimensionless variables $x = H_0 D_A/c$ and $p = dx/dz$. The dynamical variables $(x(z),p(z))$ can be integrated for a known set of equation of state parameters for different matter/energy components. However, to avoid any preference for specific dark energy models, we adopt a cosmographic approach. We consider two scenarios where the Luminosity distance is expanded as Pad\'e rational approximants using expansion in terms of $z$ and $(1+z)^{1/2}$ respectively. However, instead of directly using the Pad\'e ratios to fit kinematic quantities with data, we adopt an alternative approach where the evolution of the cold dark matter sector is incorporated in our analysis through a semi-cosmographic equation of state, which is then, used to solve the dynamical problem in the phase space. The semi-cosmographic $(D_A(z), H(z))$, thus obtained, is fitted with BAO data from DESI DR1, cosmic chronometer (CC) data and SNIa data from Pantheon+ respectively. We also consider a futuristic 21-cm intensity mapping experiment for error projections. We further use the semi-cosmographic fitting to reconstruct some diagnostics of background cosmology and compare our results for the two scenarios of Pad\'e expansions. | [
{
"version": "v1",
"created": "Wed, 05 Mar 2025 09:13:47 GMT"
},
{
"version": "v2",
"created": "Mon, 10 Mar 2025 11:30:24 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 12:09:15 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Chavan",
"Pankaj",
""
],
[
"Sarkar",
"Tapomoy Guha",
""
],
[
"Dash",
"Chandrachud B. V.",
""
],
[
"Sen",
"Anjan A",
""
]
] |
2503.03649 | Andrei Ermolaev | Andrei V. Ermolaev, Mathilde Hary, Lev Leybov, Piotr Ryczkowski, Anas Skalli, Daniel Brunner, Go\"ery Genty, and John M. Dudley | Limits of nonlinear and dispersive fiber propagation for an optical fiber-based extreme learning machine | 24 pages, 12 figures | null | 10.1364/OL.562186 | null | physics.optics cs.LG | http://creativecommons.org/licenses/by/4.0/ | We report a generalized nonlinear Schr\"odinger equation simulation model of an extreme learning machine (ELM) based on optical fiber propagation. Using the MNIST handwritten digit dataset as a benchmark, we study how accuracy depends on propagation dynamics, as well as parameters governing spectral encoding, readout, and noise. For this dataset and with quantum noise limited input, test accuracies of : over 91% and 93% are found for propagation in the anomalous and normal dispersion regimes respectively. Our results also suggest that quantum noise on the input pulses introduces an intrinsic penalty to ELM performance. | [
{
"version": "v1",
"created": "Wed, 05 Mar 2025 16:25:58 GMT"
},
{
"version": "v2",
"created": "Fri, 14 Mar 2025 09:36:47 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 16:59:36 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Ermolaev",
"Andrei V.",
""
],
[
"Hary",
"Mathilde",
""
],
[
"Leybov",
"Lev",
""
],
[
"Ryczkowski",
"Piotr",
""
],
[
"Skalli",
"Anas",
""
],
[
"Brunner",
"Daniel",
""
],
[
"Genty",
"Goëry",
""
],
[
"Dudley",
"John M.",
""
]
] |
2503.03871 | Navid Constantinou | Gregory L. Wagner and Navid C. Constantinou | Phenomenology of decaying turbulence beneath surface waves | 11 pages, 4 figures | null | null | null | physics.flu-dyn physics.ao-ph | http://creativecommons.org/licenses/by/4.0/ | This paper explores decaying turbulence beneath surface waves that is initially isotropic and shear-free. We start by presenting phenomenology revealed by wave-averaged numerical simulations: an accumulation of angular momentum in coherent vortices perpendicular to the direction of wave propagation, suppression of kinetic energy dissipation, and the development of depth-alternating jets. We interpret these features through an analogy with rotating turbulence (Holm 1996), wherein the curl of the Stokes drift, $\boldsymbol{\nabla}\times\boldsymbol{u}^S$, takes on the role of the background vorticity (for example, $(f_0 + \beta y) \hat{\boldsymbol{z}}$ on the beta plane). We pursue this thread further by showing that a two-equation model proposed by Bardina et al. (1985) for rotating turbulence reproduces the simulated evolution of volume-integrated kinetic energy. This success of the two-equation model -- which explicitly parametrizes wave-driven suppression of kinetic energy dissipation -- carries implications for modeling turbulent mixing in the ocean surface boundary layer. We conclude with a discussion about a wave-averaged analogue of the Rossby number appearing in the two-equation model, which we term the "pseudovorticity number" after the pseudovorticity $\boldsymbol{\nabla}\times\boldsymbol{u}^S$. The pseudovorticity number is related to the Langmuir number in an integral sense. | [
{
"version": "v1",
"created": "Wed, 05 Mar 2025 20:04:51 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 00:54:37 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Wagner",
"Gregory L.",
""
],
[
"Constantinou",
"Navid C.",
""
]
] |
2503.04028 | Matteo Baggioli | Jimin Bai, Long-Zhou Huang, Jin Shang, Yun-Jiang Wang, Jie Zhang, Matteo Baggioli | Stress-stress correlations in two-dimensional amorphous and crystalline solids | v2: minor revisions, matching the published version | null | null | null | cond-mat.soft cond-mat.mtrl-sci cond-mat.stat-mech | http://creativecommons.org/licenses/by/4.0/ | Stress-stress correlations in crystalline solids with long-range order can be straightforwardly derived using elasticity theory. In contrast, the `emergent elasticity' of amorphous solids, rigid materials characterized by an underlying disordered structure, defies direct explanation within traditional theoretical frameworks. To address this challenge, tensor gauge theories have been recently proposed as a promising approach to describe the emergent elasticity of disordered solids and predict their stress-stress correlations. In this work, we revisit this problem in two-dimensional amorphous and crystalline solids by employing a canonical elasticity theory approach, supported by experimental and simulation data. We demonstrate that, with respect to static stress-stress correlations, the response of a 2D disordered solid is indistinguishable from that of a 2D isotropic crystalline solid and it is well predicted by vanilla elasticity theory. Moreover, we show that the presence of pinch-point singularities in the stress response is not an exclusive feature of amorphous solids. Our results confirm previous observations about the universal character of static stress-stress correlations in crystalline and amorphous packings. | [
{
"version": "v1",
"created": "Thu, 06 Mar 2025 02:25:01 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 02:16:34 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Bai",
"Jimin",
""
],
[
"Huang",
"Long-Zhou",
""
],
[
"Shang",
"Jin",
""
],
[
"Wang",
"Yun-Jiang",
""
],
[
"Zhang",
"Jie",
""
],
[
"Baggioli",
"Matteo",
""
]
] |
2503.04179 | Taira Kawamura | Taira Kawamura, Yoji Ohashi | Engineering nonequilibrium superconducting phases in a voltage-driven superconductor under an external magnetic field | 12 pages, 8 figures | null | 10.1002/andp.202500102 | null | cond-mat.supr-con | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We theoretically investigate nonequilibrium properties of a normal metal-superconductor-normal metal (NSN) junction under an external magnetic field. When a bias voltage is applied between the normal-metal leads, the confined superconductor is driven out of equilibrium, resulting in a nonequilibrium quasiparticle distribution function having a two-step structure. Using the nonequilibrium Green's function technique, we determine a comprehensive phase diagram of the nonequilibrium superconductor. Our analysis reveals that the interplay between Zeeman-split energy bands and the nonequilibrium distribution function gives rise to a rich phase structure. Notably, we find that superconductivity destroyed by a strong external magnetic field revives by applying the bias voltage. This reentrant phenomenon is shown to originate from four effective "Fermi surfaces" that result from the combination of Zeeman-split energy bands and the two-step structure in the nonequilibrium distribution function. Our results demonstrate the possibility of controlling quantum states of matter through the combined engineering of energy band structures and distribution functions. | [
{
"version": "v1",
"created": "Thu, 06 Mar 2025 07:47:15 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Kawamura",
"Taira",
""
],
[
"Ohashi",
"Yoji",
""
]
] |
2503.04399 | Yuehan Xu Doctor | Yuehan Xu, Qijun Zhang, Junpeng Zhang, Xiaojuan Liao, Ziyi Shen, Xu Liu, Beibei Zhang, Zicong Tan, Zehao Zhou, Jisheng Dai, Xueqin Jiang, Peng Huang, Tao Wang, and Guihua Zeng | A polychromatic continuous-variable quantum communication network enabled by optical frequency combs | null | null | null | null | quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In classical communication, the introduction of polychromatic resources has rapidly boosted classical networks' rate and scale. Quantum communication is now at a similar critical stage in its development, and therefore, it is essential to investigate polychromatic quantum communication networks. In this letter, we report a polychromatic continuous-variable quantum communication network enabled by optical frequency combs. The multi-mode density matrices constituted by polychromatic quantum networks are studied. Considering the limited mode isolation, the maximum amount of information that eavesdroppers can obtain is recalculated, therefore, the total secret key rate is provided. We have also demonstrated that, compared to other multiplexing techniques, polychromatic quantum networks can theoretically achieve a secret key rate without decreasing with the increase in users. In the experiment, direct-transmission type and round-trip type quantum communication networks were built using optical frequency combs and dual-comb interference detection technology. The Gaussian-modulated continuous-variable quantum key distribution (CV-QKD) protocol has been validated, with a network capacity of 19 and a total secret key rate of 8.75 Gbps at a uniform distance of 5 km (asymptotic case), 0.82 Mbps at 120 km (finite-size effect), 89.10 Mbps at 40 km (compsable security), 13.66 Mbps at 40 km (compsable finite-size security). This implementation not only provides technical support for a high-speed multi-node quantum network, but also provides a solution for the future quantum Internet with continuous variables. | [
{
"version": "v1",
"created": "Thu, 06 Mar 2025 12:53:58 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 10:59:08 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Xu",
"Yuehan",
""
],
[
"Zhang",
"Qijun",
""
],
[
"Zhang",
"Junpeng",
""
],
[
"Liao",
"Xiaojuan",
""
],
[
"Shen",
"Ziyi",
""
],
[
"Liu",
"Xu",
""
],
[
"Zhang",
"Beibei",
""
],
[
"Tan",
"Zicong",
""
],
[
"Zhou",
"Zehao",
""
],
[
"Dai",
"Jisheng",
""
],
[
"Jiang",
"Xueqin",
""
],
[
"Huang",
"Peng",
""
],
[
"Wang",
"Tao",
""
],
[
"Zeng",
"Guihua",
""
]
] |
2503.04459 | Hongyeob Kim | Hongyeob Kim, Inyoung Jung, Dayoon Suh, Youjia Zhang, Sangmin Lee, Sungeun Hong | Question-Aware Gaussian Experts for Audio-Visual Question Answering | CVPR 2025. Code is available at https://github.com/AIM-SKKU/QA-TIGER | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Audio-Visual Question Answering (AVQA) requires not only question-based multimodal reasoning but also precise temporal grounding to capture subtle dynamics for accurate prediction. However, existing methods mainly use question information implicitly, limiting focus on question-specific details. Furthermore, most studies rely on uniform frame sampling, which can miss key question-relevant frames. Although recent Top-K frame selection methods aim to address this, their discrete nature still overlooks fine-grained temporal details. This paper proposes QA-TIGER, a novel framework that explicitly incorporates question information and models continuous temporal dynamics. Our key idea is to use Gaussian-based modeling to adaptively focus on both consecutive and non-consecutive frames based on the question, while explicitly injecting question information and applying progressive refinement. We leverage a Mixture of Experts (MoE) to flexibly implement multiple Gaussian models, activating temporal experts specifically tailored to the question. Extensive experiments on multiple AVQA benchmarks show that QA-TIGER consistently achieves state-of-the-art performance. Code is available at https://aim-skku.github.io/QA-TIGER/ | [
{
"version": "v1",
"created": "Thu, 06 Mar 2025 14:11:46 GMT"
},
{
"version": "v2",
"created": "Fri, 07 Mar 2025 09:27:33 GMT"
},
{
"version": "v3",
"created": "Wed, 11 Jun 2025 12:30:39 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Kim",
"Hongyeob",
""
],
[
"Jung",
"Inyoung",
""
],
[
"Suh",
"Dayoon",
""
],
[
"Zhang",
"Youjia",
""
],
[
"Lee",
"Sangmin",
""
],
[
"Hong",
"Sungeun",
""
]
] |
2503.04793 | Wenjie Qiu | Wenjie Qiu, Yi-Chen Li, Xuqin Zhang, Tianyi Zhang, Yihang Zhang, Zongzhang Zhang, Yang Yu | Sentence-level Reward Model can Generalize Better for Aligning LLM from Human Preference | null | null | null | null | cs.CL cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Learning reward models from human preference datasets and subsequently optimizing language models via reinforcement learning has emerged as a fundamental paradigm for aligning LLMs with human preferences. The performance of the reward model plays a crucial role in the effectiveness of alignment. Previous reward models operate at a coarse-grained level, requiring the generation of a complete response to obtain a reward value. The sparse reward may present challenges for downstream reinforcement learning. While recent efforts have attempted to learn token-level reward models, the lack of explicit semantic information makes it difficult to model the credit of every individual token. In this paper, we propose assigning scores to every sentence, introducing an intermediate-grained reward model. By segmenting the complete response into sentences and applying differential operations to reward output at the start and end positions of each sentence, we can effectively model the rewards of sentences. Moreover, a novel attention mechanism is introduced to aggregate the scores of all sentences into a response-level score, which allows it to be trained using the Bradley-Terry model. On common benchmarks, our method outperforms the response-level reward model by 2.7% on RewardBench (for reward modeling evaluation) and surpasses all baselines on AlpacaEval (for alignment evaluation). | [
{
"version": "v1",
"created": "Sat, 01 Mar 2025 14:11:04 GMT"
},
{
"version": "v2",
"created": "Sun, 08 Jun 2025 15:02:37 GMT"
},
{
"version": "v3",
"created": "Tue, 10 Jun 2025 02:25:07 GMT"
},
{
"version": "v4",
"created": "Wed, 11 Jun 2025 02:42:58 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Qiu",
"Wenjie",
""
],
[
"Li",
"Yi-Chen",
""
],
[
"Zhang",
"Xuqin",
""
],
[
"Zhang",
"Tianyi",
""
],
[
"Zhang",
"Yihang",
""
],
[
"Zhang",
"Zongzhang",
""
],
[
"Yu",
"Yang",
""
]
] |
2503.05483 | George Mihailescu | George Mihailescu, Saubhik Sarkar, Abolfazl Bayat, Steve Campbell, Andrew K. Mitchell | Metrological symmetries in singular quantum multi-parameter estimation | 24 pages with 8 figures | null | null | null | quant-ph cond-mat.mes-hall | http://creativecommons.org/licenses/by/4.0/ | The theoretical foundation of quantum sensing is rooted in the Cram\'er-Rao formalism, which establishes quantitative precision bounds for a given quantum probe. In many practical scenarios, where more than one parameter is unknown, the multi-parameter Cram\'er-Rao bound (CRB) applies. Since this is a matrix inequality involving the inverse of the quantum Fisher information matrix (QFIM), the formalism breaks down when the QFIM is singular. In this paper, we examine the physical origins of such singularities, showing that they result from an over-parametrization on the metrological level. This is itself caused by emergent metrological symmetries, whereby the same set of measurement outcomes are obtained for different combinations of system parameters. Although the number of effective parameters is equal to the number of non-zero QFIM eigenvalues, the Cram\'er-Rao formalism typically does not provide information about the effective parameter encoding. Instead, we demonstrate through a series of concrete examples that Bayesian estimation can provide deep insights. In particular, the metrological symmetries appear in the Bayesian posterior distribution as lines of persistent likelihood running through the space of unknown parameters. These lines are contour lines of the effective parameters which, through suitable parameter transformations, can be estimated and follow their own effective CRBs. | [
{
"version": "v1",
"created": "Fri, 07 Mar 2025 14:52:40 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 13:55:27 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Mihailescu",
"George",
""
],
[
"Sarkar",
"Saubhik",
""
],
[
"Bayat",
"Abolfazl",
""
],
[
"Campbell",
"Steve",
""
],
[
"Mitchell",
"Andrew K.",
""
]
] |
2503.06928 | Yanlong Wang | Yanlong Wang, Jian Xu, Tiantian Gao, Hongkang Zhang, Shao-Lun Huang, Danny Dongning Sun, Xiao-Ping Zhang | FinTSBridge: A New Evaluation Suite for Real-world Financial Prediction with Advanced Time Series Models | ICLR 2025 Workshop Advances in Financial AI | null | null | null | cs.LG q-fin.TR | http://creativecommons.org/licenses/by/4.0/ | Despite the growing attention to time series forecasting in recent years, many studies have proposed various solutions to address the challenges encountered in time series prediction, aiming to improve forecasting performance. However, effectively applying these time series forecasting models to the field of financial asset pricing remains a challenging issue. There is still a need for a bridge to connect cutting-edge time series forecasting models with financial asset pricing. To bridge this gap, we have undertaken the following efforts: 1) We constructed three datasets from the financial domain; 2) We selected over ten time series forecasting models from recent studies and validated their performance in financial time series; 3) We developed new metrics, msIC and msIR, in addition to MSE and MAE, to showcase the time series correlation captured by the models; 4) We designed financial-specific tasks for these three datasets and assessed the practical performance and application potential of these forecasting models in important financial problems. We hope the developed new evaluation suite, FinTSBridge, can provide valuable insights into the effectiveness and robustness of advanced forecasting models in finanical domains. | [
{
"version": "v1",
"created": "Mon, 10 Mar 2025 05:19:13 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 15:26:46 GMT"
}
] | 2025-06-12T00:00:00 | [
[
"Wang",
"Yanlong",
""
],
[
"Xu",
"Jian",
""
],
[
"Gao",
"Tiantian",
""
],
[
"Zhang",
"Hongkang",
""
],
[
"Huang",
"Shao-Lun",
""
],
[
"Sun",
"Danny Dongning",
""
],
[
"Zhang",
"Xiao-Ping",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.