The paper introduces an operator-informed initialization for Fourier Feature Physics-Informed Neural Networks (PINNs) to mitigate spectral bias, where certain frequencies converge slower. By analyzing training dynamics in the Neural Tangent Kernel regime, the authors derive an evolution equation linking convergence rates to the product of the differential operator’s symbol and the spectral density of initialization weights. They propose tailoring the initial weight distribution to the specific PDE, which balances convergence across frequencies and improves prediction accuracy without extra training cost, as confirmed by experiments on linear and nonlinear PDEs.
By Juan Molina, Paris Perdikaris, Mircea Petrache, Mat\'ias Courdurier, Francisco Sahli Costabal
A new framework for Bayesian data assimilation is presented that leverages pretrained stochastic interpolants, flow matching, and diffusion models as posterior samplers without retraining. By conditioning the interpolant path on observations, a shared likelihood‑score correction is applied to the drift or velocity, unifying stochastic and deterministic sampling methods. The approach yields exact posterior samples when the intermediate likelihood score is known, and practical approximations use a Gaussian surrogate with bias‑corrected mean and inflated covariance, enabling tractable high‑dimensional applications such as linear‑Gaussian dynamics, stochastic Navier‑Stokes, and urban airflow simulations with up to $O(10^4)$ degrees of freedom.
By Nikolaj T. M\"ucke, Benjamin Sanderse
PLCWorld is a closed‑loop execution environment and benchmark that evaluates large language model (LLM)–generated programmable logic controller (PLC) programs by coupling Structured Text (ST) execution with simulated plant responses and sensor feedback. It includes 100 synthetic tasks and 473 task‑condition pairs across Motion Control and Material Handling, with difficulty levels based on control‑dependency scope. The benchmark reports Task Success and Safety Violation separately and validates results through practitioner review, reference programs, counterexamples, and comparisons with independent ST runtimes, revealing performance differences among six LLMs and four generation‑and‑verification workflows.
By Yunji Kim, Yunseok Lee, Hyunwoo Seo, Jaerim Choi, Woojin Lee
TRACE is a reproducible benchmark for electricity price forecasting that pairs 7,300 zone–day instances from five U.S. market zones with official operational text available at the forecast cutoff, ensuring no post‑cutoff information leakage. The benchmark evaluates semantic alignment and forecasting value, showing that the text aligns with central movement and tail risks in ground‑truth prices, especially for upper‑tail risk. Forecasting models using TRACE’s text achieve a median 7.4% reduction in upper‑tail pinball loss, while a controlled ablation that mismatches text reverses these gains.
By Xinyi Yi, Moy Yuan, Ioannis Lestas
The paper introduces Calibrated User Embeddings (CUE), a framework that encodes real user sessions into continuous representations and decodes them into persona commands to steer large language models (LLMs) as user simulators without additional training. CUE enables user-conditioned replay of past interactions and generates novel personas that better align with real-user success rates and failure patterns. Evaluations on the τ²‑Bench dataset show that CUE‑driven simulators commit fewer simulator‑attributed errors, more accurately reproduce real‑user failure modes, and maintain competitive user fidelity across various tasks and LLMs.
By Anjali Kantharuban, Jonas Mueller
The paper introduces a two‑stage framework for auditing false alarms in safety monitors for language‑model agents, treating the problem as a positive‑unlabeled (PU) ranking task. It adapts safe references toward the alarm domain, consolidates ordering from multiple PU models, and refines the ranking with structural guidance, all without requiring alarm safety labels. The method improves macro AUPRC to 0.6444 and recovers 33.3% more false alarms at a 5% review budget compared to leading PU baselines.
By Xichen Yan, Chongyang Gao, Kezhen Chen, Guangyi Zhang, Jiaqi Wu, Lixu Wang
RNADynBench is a new standardized benchmark for RNA molecular dynamics, comprising 2,585 high‑quality 100‑nanosecond all‑atom trajectories with carefully controlled data splits. RNADynNet is a unified model that simultaneously generates RNA trajectories and extracts dynamics fingerprints from a single conformer, leveraging coordinate denoising, alignment, and physical grounding. The model achieves strong agreement with MD data, achieving RMSF correlations of 0.875 and 0.766 on test sets, including a high‑flexibility challenge set.
By Yiming Huang, Lennart Bastian, Hanqun Cao, Luis Vollmers, Tolga Birdal
PoCoFL is a federated learning framework that separates the type of FL, policy semantics, and cryptographic realization to ensure policy compliance. It formalizes client and aggregation requirements as policy-dependent relations, allowing clients to prove compliance with commitments and zero-knowledge proofs, and aggregators to prove correct processing of admitted contributions. The paper demonstrates PoCoFL through four instantiations—vanilla, continual, personalized, and threshold-encrypted FL—showing its versatility and practical feasibility across different learning objectives.
By Dominik Roy George, Varesh Mishra, Aysajan Abidin
The paper investigates whether clinical concepts are represented as distinct, causally influential centers within the latent space of large language models (LLMs). By evaluating eleven open-weight LLMs, the authors discover that each model contains dedicated clinical concept centers that are interpretable, activate only on relevant clinical narratives, and drive model behavior in both constrained and open-ended contexts. These centers can be leveraged for evaluation and performance improvement, as steering models along them enhances downstream clinical outcomes and aligns with clinician preferences.
By Aishik Nagar, Abhishek Vaidyanathan, Arun-Kumar Kaliya-Perumal, Elijah Tzen Hsuen Boey, Stefan Winkler
CuBEs introduces culturally‑situated behavioral evaluations for large language models, adding cultural context to test scenarios and assessments. The authors built a human‑labeled dataset covering 12 cultures, revealing significant cross‑cultural differences that one‑size‑fits‑all judgments miss. Evaluating 13 LLMs shows that culturally situated tests uncover varied behaviors, such as Western political bias versus non‑Western religious or colonial biases, which standard evaluations overlook.
By Hoda Ayad, Tanu Mitra, Abhishek Mukherji
TasteBench is a multimodal benchmark designed to accelerate sustainable protein discovery by providing computational proxies for sensory prediction. It includes a food-level ranking task based on over 21,000 human evaluations of 215 plant-based foods across 24 categories, and a molecular-level taste classification task covering 15,000 flavor molecules. The benchmark offers baseline models, characterizes inter-rater agreement and reliability limits, and demonstrates that the best model achieves pairwise accuracy comparable to individual human panelists.
By Anna T. Thomas, Sohum Patnaik, Caroline Cotto, Benjamin Sanchez-Lengeling
VisionMX introduces a post‑training microscaling (MX) quantization technique for vision models, addressing three key error sources identified in direct conversion: block‑scale representation, misalignment of small convolutional weight tensors, and underuse of signed codes for nonnegative activations. The method optimizes bounded weight rounding and incorporates a foldable affine correction for activations, improving performance across image classification, object detection, semantic segmentation, and low‑light image enhancement tasks. VisionMX outperforms direct conversion and other post‑training quantization baselines, especially in architectures most sensitive to MX conversion.
By Elad Dror Cohen, Ofir Gordon, Lior Dikstein, Idan Achituve, Hai Victor Habi
The paper introduces PI‑SME, a Path‑Integral Surrogate Model Extension, to improve gradient inversion attacks in federated learning. By treating a client’s accumulated update as a path integral of the gradient field and approximating it with Gauss–Legendre quadrature over a learnable Bézier path, PI‑SME reconstructs private input images more accurately than existing surrogate baselines. Experiments on CIFAR‑100 and FEMNIST demonstrate its superior performance across various trajectory lengths and class‑restricted batches.
By Agnivo Ghosh, Saumik Bhattacharya
The paper introduces LatticeSMC, a sampler for chunked sequence generators that balances inference-time compute across chunk indices and denoising steps using a Feynman‑Kac model on a two‑dimensional lattice. By resampling at chunk boundaries and, when scoring is free, within chunks, it achieves higher beat alignment and prompt adherence on music‑to‑dance diffusion and text‑to‑music generation while maintaining quality. The method also outperforms existing steering techniques on long‑range rewards and is preferred by human raters in most pairwise comparisons.
By Xuanchen Wang, Heng Wang, Weidong Cai
HakemBench is a Turkish benchmark for typed decision-making tasks, comprising 2,346 items and 4,275 choice, yes/no, and score questions across seven tracks (fact‑check triage, education, guardrails, legal routing, moderation, spam and phishing, and customer support). The benchmark evaluates models on decision quality (macro F1), calibration (normalised Brier score), and selective automation (normalised area under the generalised risk‑coverage curve), combining these metrics by a geometric mean and reporting confidence intervals from 2,000 bootstrap draws. Gold labels are generated from blind passes of a single AI model family compared with votes from other large language model families, and the benchmark’s leader achieved a composite score of 0.888 while the lab’s own model ranked 7th with 0.660.
By Sait Furkan Teke (ufak AI)
The paper introduces AURA, an LLM-powered mask‑reconstruct framework for anonymizing text while preserving utility. It decouples privacy localization from utility‑preserving reconstruction and uses adversarial checks to select candidates. Experiments on real‑user interview transcripts show that AURA achieves the lowest agentic re‑identification rates among non‑DP methods and retains more contextual utility than prior LLM anonymizers.
By Ziwen Li, Jianing Wen, Tianshi Li
The paper introduces Causal-Invariant Masking (CIM) to better quantify epistemic uncertainty in Multimodal Large Language Models (MLLMs) by measuring semantic shift between original predictions and those conditioned on a causally-focused view. It proposes Semantic Divergence as a core metric that converges to the variance of the model’s sensitivity to non‑causal correlations, and introduces Expected Embedding Drift (EED) as a fast geometric proxy that estimates this shift directly in the embedding space. Experiments demonstrate state‑of‑the‑art uncertainty quantification performance and a nearly 50% speedup with EED.
By Haoyang Luo, Linwei Tao, Jie Gui, Xinghao Chen, Chang Xu, Jianyuan Guo, Minjing Dong
The paper examines the limitations of global unsafety signals used in training‑free safeguards for text‑to‑image generation, showing a trade‑off between coverage and selectivity. It introduces CALM, a counterfactual adaptive local modulation method that locally corrects unsafe token representations while preserving benign content. Experiments demonstrate that CALM outperforms global removal approaches in suppressing unsafe content without compromising utility.
By NaHyeon Park, Minhyun Lee, Hyunjung Shim
Enrich-on-Graph (EoG) is a flexible framework that uses large language models to enrich knowledge graphs, thereby bridging the semantic gap between structured graphs and unstructured queries in complex reasoning tasks. By leveraging LLMs’ prior knowledge, EoG enables efficient evidence extraction from knowledge graphs, achieving precise and robust reasoning while maintaining low computational costs and scalability. The authors also introduce three graph quality evaluation metrics for query‑graph alignment, theoretically validate their optimization objectives, and demonstrate state‑of‑the‑art performance on two KGQA benchmark datasets.
By Songze Li, Zhiqiang Liu, Zhengke Gui, Huajun Chen, Wen Zhang
The paper examines a trace‑completeness assumption in language‑enabled robot systems that combine semantic‑graph planning with temporal‑logic safety monitors. By auditing RoboGuard, the authors compare the monitor’s verdict on a surface plan with its verdict on a graph‑refined trace under the same LTL specification. Their evaluation of 28 controlled cases and 14 end‑to‑end scenarios shows that all targeted abstraction cases exhibit the predicted surface‑versus‑refined discrepancy, while controls behave as expected, suggesting that graph‑based trace refinement can serve as a lightweight mitigation and diagnostic tool for physical‑AI safety monitors.
By Stabak Das, Priyesh Ranjan, Xiangfang Li, Lijun Qian