Leaderboards, eval harnesses and ablations — the contested business of deciding which model is actually better.
In reinforcement learning, advantage-based value-learning methods, unlike standard Q-learning, share information across actions by decomposing the action-value function into a state-dependent value co...
Large language model (LLM) agents now fix software, operate computers and serve customers, but their harnesses, the programs that run them, call a large LLM at every step, which makes long tasks slow...
Resting-state functional connectivity is widely used in machine learning-based classification of neuropsychiatric disorders. Quantum machine learning models have been proposed for functional connectiv...
Large language models (LLMs) are increasingly used for election information, yet little is known about how external web content shapes the political information they present. Generative engine optimiz...
Vision-language-action (VLA) models are expected to follow instructions like their language backbones, yet prior work has focused on What-level instructions that specify the task rather than How-level...
Reinforcement learning with verifiable rewards (RLVR) math corpora are routinely screened for benchmark overlap, but the reliability and data-loss costs of cleaning policies remain underexamined. We a...
TextualRL is a gradient‑free reinforcement learning framework that updates the textual context guiding large language model agents without altering their parameters. It treats groups of rollouts from the same task as units of critique, contrasting successful and failed runs to generate context patches, and validates these patches using held‑out rewards. Experiments on six benchmarks show that TextualRL boosts test accuracy by 24.4 and 15.3 percentage points for GPT‑5.5 and Qwen3.8‑27B, respectively.
The paper introduces GREAT, a Graph Edge Attention Network that operates directly on edge space to handle asymmetric energy costs in the Electric Vehicle Routing Problem (EVRP). Integrated into a reinforcement learning framework, GREAT jointly optimizes routing and charging decisions, outperforming a strong HGS-based heuristic by up to 8.78% in energy savings and reducing runtime by a factor of 60. It also shows competitive performance against a custom adaptive large neighborhood search and achieves state‑of‑the‑art results on other asymmetric routing benchmarks such as TSP, CVRP, and Orienteering.
The paper surveys open, non‑proprietary large language models (LLMs) and AI foundation models (AIFMs) for clinical use, presenting a taxonomy of current state‑of‑the‑art models. It empirically tests open‑weight LLMs on a personalized‑prescription task, finding that they match or surpass a proprietary baseline in overall suitability accuracy, especially when combined with Retrieval‑Augmented Generation (RAG). A subjective evaluation by blinded raters shows mixed effects of RAG on perceived suitability across multiple dimensions, and the authors discuss ethical risks and the need for responsible deployment.
The paper reviews recent work on natural language generation from visual events, emphasizing that tasks involving sequences of images or video frames are fundamentally about modeling complex relationships between evolving visual events and the language used to describe them. It surveys five related tasks, examines common modeling and evaluation methods, and highlights shared challenges across these tasks. The authors then outline key open questions and suggest future research directions to address these challenges.
DiffPTS rethinks the evidence lower bound (ELBO) for diffusion-based probabilistic time‑series forecasting by adopting a Location‑Scale Noise Model, which yields a Gaussian negative log‑likelihood objective and a unified joint training objective. The framework enables end‑to‑end optimization of all components within the ELBO, unlike prior methods that treat estimator training as separate regression tasks. On multiple benchmarks, DiffPTS achieves state‑of‑the‑art performance, reducing CRPS/MSE by over 14.53%/16.55% compared to existing diffusion‑based models.
EMG-FM-Bench is a benchmark that unifies 20 public electromyography datasets, totaling over one million EMG segments, to evaluate how foundation models transfer and adapt to EMG tasks. It tests nine pretrained models across four scenarios: frozen versus fully fine‑tuned performance, the benefit of pretraining versus training from scratch, generalization to new users with limited data, and performance across different EMG tasks. Results show that linear probing reveals useful pretrained representations, full fine‑tuning can significantly alter performance, pretraining benefits vary by model, performance drops on new users, and few‑shot adaptation improves macro‑F1 in most cases but only partially recovers lost performance.
The paper investigates whether large language models (LLMs) can learn from rewards presented in context, a process termed in-context reinforcement learning (ICRL). Experiments across four benchmarks and six models show that while LLMs read reward signals, altering or removing these rewards has little impact on performance, and shuffled or corrupted trajectories are as effective as real ones. These findings suggest that direct ICRL behaves more like in-context learning (ICL) than true inference-time reinforcement learning.
The paper introduces AutoConvexRelax, a learning-based framework that automates the design of convex relaxations for nonconvex optimization problems. By representing a problem as a tripartite graph of variables, symbolic terms, and objective/constraint positions, the method learns a policy to compose symbolic rewrites that generate valid relaxation trajectories. Experiments on quadratic and fractional problems show that the learned policy improves McCormick lower bounds by 13.52% and achieves near‑semidefinite bounds while using smaller formulations.
The paper investigates whether accuracy and order sensitivity in large language model (LLM) multiple‑choice question (MCQ) evaluation move together. Using proprietary and open‑weight models on MMLU and ARC‑Challenge, the authors measure order sensitivity with flip rate and recall standard deviation (RStd) and find that these metrics often diverge from accuracy and from each other. They conclude that accuracy alone is insufficient and recommend reporting both flip rate and RStd alongside accuracy.
Spectral-MoE introduces a method for allocating heterogeneous per-layer expert utilization in dense-to-mixture-of-experts (MoE) conversion of large language models. It offers both static and dynamic allocators that learn to distribute a utilization budget across layers, improving performance over uniform allocation in both analytic- and learning-based conversion baselines. Experiments on Llama-2-7B, Llama-3-8B, and Qwen2.5-7B show significant reductions in perplexity and accuracy gains at various FFN utilizations.
arXiv:2610.19165v1 Announce Type: cross Abstract: Many real-world questions demand not a single fact but an organized collection of them, yet existing factual knowledge benchmarks almost exclusively...
AromaGen is a 12‑channel wearable system that translates free‑form natural‑language descriptions into olfactory experiences by selecting from 12 base odorants that span a semantically derived olfactory space. The system supports iterative refinement via natural‑language feedback. In a study with 60 participants and 50 benchmark smells, users could identify AromaGen‑generated scents above chance, and AI‑generated compositions performed comparably to human experts, with feedback improving perceived similarity.
The paper introduces Verifiable Agent Insurance (VAI), a framework that lets AI agent providers commit to policy compliance through smart contracts and escrow funds on a blockchain. When a user detects a policy violation, a verifier reviews execution traces logged by the agent harness; if the violation is confirmed, the escrow is paid to the user. Experiments on three agent benchmarks show VAI detects 42.2% of violations, rising to 83.0% when agents emit explicit structured records.
The paper investigates the projector training stage in multimodal large language models, showing that only a small fraction of data is needed and that projectors from different datasets can be interpolated with little loss. It proposes PORTAL, a training‑free initialization that uses optimal transport between Gaussian approximations of vision and language features, eliminating the need for paired image‑caption data. PORTAL matches or surpasses the standard pipeline across multiple backbones, encoders, and benchmarks.