All Articles
17670 articles total
Delta-JEPA: Learning Action-Sensitive World Models via Latent Difference Decoding
Introduction of Delta-JEPA, a world model that uses Latent Difference Action Decoding to prevent representation collapse and improve action-sensitive planning in visual tasks.
Embodied CAD: Solver-Grounded LLM Agents for Parametric B-Rep Assembly Modeling
Embodied CAD presents a framework using solver-grounded LLM agents to iteratively create and repair parametric B-Rep assembly models in industrial CAD.
Spatial Reasoning via Modality Switching Between Language and Symbolic Representation
Research demonstrating that allowing LLMs to switch from natural language to symbolic grid representations can improve spatial reasoning performance by up to 42%.
Benchmarking Large Language Models on Floating-Point Error Classification
The InterFLOPBench benchmark evaluates LLMs' ability to statically detect floating-point errors in C code, showing high accuracy for simple errors but lower for subtle numerical issues.
HistoriQA-ThirdRepublic: Multi-Hop Question Answering Corpus for Historical Research, Parliamentary Debates from the French Third Republic (1870-1940)
HistoriQA-ThirdRepublic is a new French-language multi-hop QA dataset based on parliamentary debates and newspapers from the French Third Republic.
CryoACE: An Atom-centric Framework for Accurate and Automated Model Building in Cryo-EM
CryoACE introduces an atom-centric framework for automated model building in Cryo-EM, improving accuracy in reconstructing atomic graphs for heterogeneous protein structures.
Optimization Algorithms for Joint OFDM Waveform Design and RIS Configuration in 6G Networks: From Convex Relaxation to Foundation Models
A comprehensive survey of optimization algorithms for 6G network waveform design and RIS configuration, contrasting model-based methods with emerging ML and foundation models.
Smart charging of large fleets of Electric Vehicles: Independent Multi-Agent Reinforcement Learning approaches
Comparison of contextual combinatorial bandits and policy gradient algorithms for the decentralized smart charging of large electric vehicle fleets.
ReGRPO: Reflection-Augmented Policy Optimization for Tool-Using Agents
ReGRPO is a framework that enhances tool-using agents by combining Reflection-of-Thought triplets with Group Relative Policy Optimization to improve error recovery.
ArXiv's Next Chapter
A discussion thread on Hacker News regarding the future direction and evolution of the ArXiv preprint server.
Cross-Domain Feature Expansion for Tabular Medical Data via Knowledge Graphs Injection
Researchers introduce MedKGTab, a framework that uses knowledge graphs to expand tabular medical data and infer missing biomedical features.
ClawArena-Team: Benchmarking Subagent Orchestration and Dynamic Workflows in Language-Model Agents
ClawArena-Team is a new benchmark designed to measure the management ability of a single LLM acting as a leader orchestrating multiple subagents.
HealthAgentBench: A Unified Benchmark Suite of Realistic Agentic Healthcare Environments for Challenging Frontier AI Agents
Microsoft introduces HealthAgentBench, a suite of 54 realistic healthcare tasks to evaluate the reasoning and execution capabilities of frontier AI agents.
AI-Assisted Discovery of Convex Relaxations via Dual Agents
The authors use a dual-agent autoresearch paradigm to discover tighter convex relaxations for sharp-constant inequalities in optimization.
Agentic RAG-VLM: Affordance-Aware Retrieval-Augmented Generation with Self-Reflective Planning for Robotic Grasping
Agentic RAG-VLM is a framework for robotic grasping that combines affordance-aware retrieval, scene graph reasoning, and self-reflective planning.
Towards Inclusive Mobility Modeling: Characterizing and Evaluating Elderly Trajectory Patterns in Urban Systems
This study analyzes how the underrepresentation of elderly mobility patterns in urban datasets leads to bias in synthetic trajectory generation models.
Long-term Traffic Simulation via Structured Autoregressive Modeling
RosettaSim is introduced as a framework for long-term traffic simulation for autonomous driving using structured autoregressive modeling and LLMs.
Thinking Before Retrieving: Robust Zero-Shot Composed Image Retrieval via Strategic Planning and Self-Criticism
PEC-CIR is a training-free framework for zero-shot composed image retrieval that uses a Planner-Executor-Critic architecture to reduce generative errors.
Agentic-Ideation: Sample Efficient Agentic Trajectories Synthesis for Scientific Ideation Agents
Agentic-Ideation is a framework for synthesizing sample-efficient agentic trajectories to train LLMs for automated scientific ideation.
The first early human eggs from stem cells
Researchers have successfully created early human eggs from stem cells, a significant breakthrough in reproductive biology.