Wei Wang

Uncovering crucial supporting lemmas for theorem proving by traversing multi-hop dependencies in a formal library's declaration graph.

A fixed-size latent memory, explicitly optimized for robotic control via question-answering and DINO feature reconstruction, unlocks state-of-the-art long-horizon manipulation performance.

This work introduces ACLArena, a framework for comprehensively studying, analyzing, and evaluating Agent Continual Learning, and proposes a new ACL recipe that combines offline replay over high-quality trajectories with a routed network of multiple LoRA experts each specialized via RL, substantially improving the agent's ability to learn across multiple domains.

Video can be reimagined as a dynamic interplay of stable contexts and evolving events, revolutionizing real-time interaction capabilities in AI.

RetroAgent revolutionizes retrosynthesis planning by combining LLMs with structured memory, leading to significantly improved decision-making in complex chemical searches.

Current LLM-driven theorem provers fall short in addressing the complexities of frontier mathematics, necessitating a shift towards research agents that can engage in rigorous mathematical exploration.

DT-Guard outperforms larger models in safety classification while maintaining low-latency performance, proving that reasoning supervision can be efficiently internalized.

Higher resolution in real-time audio-visual interactions can be achieved without sacrificing latency, enabling clearer agent representation in conversations.

Failure-driven post-training, combined with a meticulously curated 10M token STEM dataset, unlocks a 4.68% performance boost in LLM reasoning, proving that strategic data synthesis around model weaknesses is a powerful path to improvement.

An open-source ecosystem for agentic learning, complete with a trained agent and novel policy optimization, promises to accelerate research by providing a standardized, scalable platform.