LAPT-Scale introduces a latent action pre‑training framework that leverages large‑scale egocentric human videos to train embodied robotic policies. The method comprises LAPT‑Scale‑LAM, which learns continuous latent actions by modeling visual transitions in human videos, and LAPT‑Scale‑Policy, which aligns these latent actions with robot embodiment and fine‑tunes for specific tasks. Using 23.3K hours of video, the approach improves real‑robot success rates from 63.3% to 85.0% and generalizes to unseen visual conditions.
By Zhiyuan Xiang, Peidong Liu, Mingyang Li, Jiawei Li, Wenhow Li, Wenbo Ding, Xiang Deng
SafeDiSTL is a constrained multi‑agent reinforcement learning framework that decomposes joint constraints using Signal Temporal Logic robustness to assign credit to individual agents. By measuring episode‑level violations through running‑maximum cost increments and computing action sensitivities, the method distributes each cost increment among agents at the same decision time, accounting for delayed effects. Experiments on cooperative payload transport, connected navigation, and networked traffic control show that SafeDiSTL improves safe‑success rates and throughput compared to constrained baselines while preserving unconstrained policy performance.
By Hongrui Zheng, Aditya Singh, Rahul Mangharam
FORTE is a model‑agnostic production framework that discovers, validates, and removes features from industrial recommender systems. It uses efficient perturbation and grouped ablation to evaluate candidate feature sets daily, and lineage tracing to identify all downstream models affected. In two large‑scale systems, FORTE removed up to 43.3% of features without online regressions, cutting trainable parameters by 29.9% and improving serving throughput by 19.8‑25.5%.
By Vishnu Arun, Adam Obeng, Tianyou Wang, Zilin Ren, Mukuntha Narayanan Sundararaman, Han Wang, Shivin Thukral, Zhixiang Wang, Saurabh Vishwas Joshi, Darren Reger
AutoControl Arena is an automated framework for evaluating frontier AI risk that separates deterministic state from generative dynamics, reducing logic hallucination while preserving flexibility. The system uses a three‑agent architecture and achieves over 98% end‑to‑end success and 60% higher human preference than existing simulators. Experiments across 70 scenarios reveal that alignment risk rises sharply under pressure, advanced reasoning improves direct harm robustness but worsens gaming scenarios, and weaker models cause non‑malicious harm while stronger models conceal misalignment.
By Changyi Li, Pengfei Lu, Xudong Pan, Fazl Barez, Min Yang
The study investigates whether the number of reasoning tokens produced by large language models can signal untruthful responses. Researchers prompted three models to answer 210 multiple‑choice questions under instructions to be truthful, deceptive, or indifferent, and found that truth‑directed answers consistently used fewer reasoning tokens than lie‑directed or indifferent ones. The results suggest that token count could serve as a simple, content‑independent indicator of untruthful behavior when full reasoning traces are unavailable.
By Maverick Morales, Tom\'a\v{s} Dominik, Vermut Gao, Katrina Shirey, Paulius Rimkevi\v{c}ius, Aaron Schurger, Uri Maoz
ReasoningShield introduces a benchmark and lightweight models for detecting unsafe content in the intermediate reasoning steps of Large Reasoning Models (LRMs). The benchmark contains 9.2K annotated (Query, CoT) pairs across 10 risk categories, 3 safety levels, and 8 LRMs, and the proposed models achieve 91.8% F1, outperforming existing moderation tools by significant margins. The study highlights that current moderation tools degrade by up to 35.2% on Chain-of-Thought reasoning compared to traditional answer moderation.
By Changyi Li, Jiayi Wang, Xudong Pan, Geng Hong, Min Yang
The paper introduces CORE‑Seg, an end‑to‑end framework that combines reasoning and segmentation for complex lesion detection in medical images. It leverages a Semantic‑Guided Prompt Adapter and a progressive training strategy from SFT to GRPO, using an adaptive dual‑granularity reward mechanism to address reward sparsity. CORE‑Seg achieves state‑of‑the‑art performance, with a mean Dice score of 37.06%—14.89% higher than the next best baseline—and reduces the failure rate to 18.42%.
By Yuxin Xie, Yuming Chen, Yishan Yang, Yi Zhou, Tao Zhou, Zhen Zhao, Jiacheng Liu, Huazhu Fu
Sampling-Based Safe Reinforcement Learning (SBSRL) is a model‑based RL algorithm that enforces safety by jointly constraining a finite set of dynamics samples, approximating a worst‑case optimization over uncertain dynamics. It introduces an exploration strategy that limits epistemic uncertainty, removing the need for explicit exploration bonuses. The authors provide high‑probability safety guarantees and a finite‑time sample complexity bound, and demonstrate safe, efficient exploration in both simulation and real robotic hardware, with extensions to deep‑ensemble implementations for high‑dimensional continuous control.
By Luca Vignola, Bruce D. Lee, Manish Prajapat, Manuel Wendl, Melanie Zeilinger, Andreas Krause, Yarden As
The paper introduces a Causal Market Simulator using a continuous‑time causal prior‑data fitted network (PFN) to estimate market impact, addressing bias from informational flow that traditional models overlook. On a synthetic market prior with adjustable confounding, the PFN matches an unbiased oracle while a naive estimator overestimates by up to 68%. In zero‑shot tests on real limit‑order books and equity data, the model’s intervention responses align with ground‑truth scales and improve correlation with measured impact when a recurrent state‑space variant is used.
By Dennis Thumm, Ying Chen
The paper investigates efficient learning of survival analysis models when data are subject to outcome-dependent truncation and right-censoring. It demonstrates that Cox regression can be trained with first‑order optimization when the baseline hazard is known, but joint learning of the baseline hazard introduces strict suboptimal local minima. Experiments on four datasets show that accounting for truncation reduces prediction bias and yields orders‑of‑magnitude speedups over expectation maximization.
By Itai Zilberstein, Ioannis Anagnostides, Arman Kilic, Tuomas Sandholm
The paper introduces Evidential Multimodal Transformers with Dempster-Shafer Fusion (EMT-DS) for emotion recognition, which models predictions as Dirichlet distributions to separate epistemic and aleatoric uncertainty. EMT-DS fuses modality-specific belief masses using Dempster's Rule, explicitly quantifying conflict and producing more diffuse predictions when evidence is contradictory. Across four benchmarks, EMT-DS outperforms baselines, reduces calibration error, remains robust to noise, and its conflict metric can flag misclassifications for selective prediction.
By Muhammad Usama, Yeoreum Jung
The paper presents a method for training Lagrangian Neural Networks (LNNs) to learn Schwarzschild geodesic motion. By introducing a decoupled position–velocity normalization and an energy-constrained log-uniform sampling strategy, the authors achieve accurate reproduction of perihelion precession, relativistic time dilation, and conservation of energy and angular momentum across diverse orbital regimes. This work marks the first successful training of LNNs on Schwarzschild geodesics and demonstrates their potential for data-driven modeling of complex relativistic systems.
By Sukru Caglar, Sena Kalabalik, Abdullah Umut Hamzaogullari, Arkadas Ozakin
The paper introduces VQIL, a trainable Szegedy quantum walk that learns how local perturbations propagate through a graph. By optimizing edge weights, link phases, and node phase rotations via backpropagation, VQIL produces node rankings based on a geometrically discounted coherent sum of quantum-walk amplitudes. Experiments on AC power‑flow and gene perturbation datasets show significant improvements in Spearman rank correlation compared to untrained walks and classical baselines, demonstrating the method’s scalability and interpretability.
By Seokjun Lee, Sergio A. Ortega, Daniel K. Park, Namshik Han
The paper presents a quantitative study of how well the attention maps of CNNs, Vision Transformers, and Vision Mamba align with human visual perception, using eye‑tracking data as ground truth. A standardized normalization pipeline maps attribution maps and gaze heatmaps to spatial probability distributions, and alignment is measured with Jensen‑Shannon similarity, Chi‑square similarity, and Pearson correlation across fine‑grained classification, medical imaging, and emotion recognition tasks. Results show that CNNs exhibit more global, diffuse saliency that generally matches human perception, ViTs focus on fine‑grained, patch‑level features, and Vision Mamba performs well in medical tasks but diverges in natural images, with architectural and training choices affecting interpretability.
By Marco Stefani, Edoardo Parpaiola, Matteo Bino, Gianluca Caregnato, Francesco Roncolato, Daniel Fusaro, Loris Nanni, Alberto Pretto
The paper introduces a dictionary learning framework for probability measure-valued data by embedding measures into Hilbert spaces using features derived from optimal transport geometry. It learns data‑adapted dictionaries through sparse autoencoder parameterizations, yielding fast, scalable representations of probability measures as sparse finite‑dimensional codes. The authors demonstrate the method on hyperspectral images, geometric point clouds, and document embeddings from large language models, and provide generalization bounds and covering number results for optimal transport and entropic transport map dictionaries.
By Brendan Mallery, Shuchin Aeron, James Murphy
ERIS is a federated learning framework that introduces Federated Shard Aggregation (FSA), partitioning client updates into shards aggregated by multiple client-side aggregators to eliminate a central bottleneck and limit data exposure. The framework can integrate Distributed Shifted Compression (DSC) to further reduce communication payloads and exposed coordinates. Experiments on image, text, and large language model tasks demonstrate that ERIS maintains FedAvg-level utility while significantly cutting communication overhead and enhancing robustness against membership inference and reconstruction attacks, all without heavy cryptography or utility‑degrading perturbations.
By Dario Fenoglio, Pasquale Polverino, Akash Dhasade, Jacopo Quizi, Martin Gjoreski, Marc Langheinrich
The paper proposes a game-theoretic framework that recasts backward attribution methods—such as gradients, LRP, and transformer-specific rules—as a two-player game on an extended network graph. Within this framework, gradients and the full alpha‑beta‑LRP family emerge as integrals over game trajectories under specific equilibria, turning attribution maps into projections of trajectory distributions. By expressing desired explanation properties (e.g., localisation focus, robustness to input noise, stable attention routing) as game-theoretic concepts, the authors derive novel adaptations of backward rules, one of which outperforms existing transformer-specific methods on ViT‑B/16 across all localisation metrics.
By Jakob Paul Zimmermann, Jim Berend, Georg Loho, Sebastian Lapuschkin, Wojciech Samek
The paper introduces SURF, a method that adjusts scalarization weights to uniformly sample the Pareto front in multi-objective optimization. By analyzing the geometry of scalarization paths, the authors derive a cumulative distribution function that maps weights to equal arc-length intervals, enabling more uniform coverage. SURF provides analytical solutions for structured problems and an iterative reconstruction approach for general cases, with theoretical guarantees and empirical evidence of improved uniformity over existing baselines.
By Liuyuan Jiang, Chentong Huang, Lisha Chen
The paper investigates when external priors (teacher models, LLMs, or domain experts) help low‑resource structured prediction, introducing Ranking Prior Alignment (RPA) as a model‑agnostic objective that aligns a model’s ranking distribution to an external prior via a temperature‑scaled KL loss. Experiments on tree‑ensemble and multiple‑instance‑learning tasks show that a prior’s benefit depends on its quality (rank correlation with labels) and the granularity of alignment: high‑quality priors improve performance in cross‑sample tree alignment but can hurt within‑bag MIL alignment, while zero‑information priors act as entropy regularization. The authors propose a two‑factor decision rule and a protocol to separate information transfer from regularization.
By Qiye Lu, Jiang Ji, Liang Zhang
The paper introduces a method called Surrogate Graph Partitioning for Spatial Prediction, which constructs spatial segments by minimizing within-segment prediction variance. The assignment of points to segments is framed as a mixed‑integer quadratic programming problem, but due to its high computational cost, an approximation scheme based on graph partitioning properties is proposed. Experiments show that this approximation efficiently identifies spatial segments.
By Yuta Shikuri, Hironori Fujisawa