A new contrastive framework called SARA aims to improve the robustness of preference-based reinforcement learning against noisy or inaccurate human labeling.
This study explores the feasibility of running modern multimodal AI models on legacy hardware by deploying a compact assistant on a 2011-era GPU.
Researchers have developed SOReL, a Bayesian model-based reinforcement learning method that enables hyperparameter selection without requiring online environment interactions.
Researchers have developed a retrieval-augmented generation framework to automate the creation of regulation-compliant test scenarios for autonomous driving simulations.
Researchers have developed a new evaluation framework called JKP to test the epistemic stability of vision-language models when subjected to repetitive adversarial questioning.
This paper presents a mathematical framework for reinforcement learning in environments where the underlying dynamics switch between different states.
The SAGA agent framework is proposed to improve long-horizon strategic planning in complex games by better managing relational structures and evolving goals.
A new study evaluates the effectiveness of belief-based memory architectures in LLM agents using Bayesian inference to manage long-term information.
A new post-training method for multimodal document question answering focuses on reasoning-free alignment to reduce computational costs and improve visual grounding.
Researchers introduced Contrastive Policy Optimization to improve reinforcement learning by using token-level disagreement to better distinguish between useful uncertainty and harmful confusion.
This research proposes a framework to transition LLM-based agents from stateless interactions to situated models capable of maintaining temporal continuity for psychological support.
A new flow-matching technique for feature-space world modeling addresses the trade-off between reconstruction quality and predictive accuracy in stochastic environments.
Researchers have developed a multi-stage agentic framework designed to improve the accuracy and reasoning capabilities of AI models in interpreting chest X-ray images.
A new cloud-edge framework combines gesture detection with LLM and VLM agents to enable efficient multimodal interaction for robots with limited onboard computing power.
This study compares the performance of leading vision-language models against human experts in predicting building typologies from street-level imagery.
A new method called SciDiagramEdit uses natural language instructions to automate the complex process of editing scientific diagrams and figures.
The teLLMe system provides a framework for performing exploratory causal analysis on observational urban traffic data derived from dashcam video.
The proposed NORACL framework introduces a neurogenesis-inspired approach to continual learning that allows models to adapt their architecture dynamically to new tasks.
The CoreForge project demonstrates an iterative workflow where large language models are used to construct a MaxSAT solver directly from academic literature rather than existing codebases.
A new benchmark called Alipay-PIBench has been introduced to evaluate the performance of AI coding agents in handling complex payment integration tasks.
The GRASP framework improves agentic retrieval-augmented generation by optimizing when to retrieve information and controlling the granularity of the retrieved context.
The authors present MorphologyFM, a foundation model designed to learn clinical waveform representations while preserving the structural morphology of ECG and pulse oximetry data.
A randomized field experiment evaluated how professional and playful emails rewritten by GPT-5 affected recipient engagement and behavior in workplace communications.
The proposed RAD framework improves offline reinforcement learning by retrieving high-quality demonstrations to help models generalize beyond static training datasets.