AgentsSci — Agents for Science

community

https://chihhsuan-yang.github.io/

AI & ML interests

Research harnesses for LLM agents, multi-agent systems, and agentic AI for Science. Open research on evaluation, feedback, memory, and reproducibility.

Recent Activity

anjir29 updated a dataset 6 days ago

AgentsSci/AAMAS2027-ConditionalMemory

UHXinyang published a dataset 7 days ago

AgentsSci/ICLR2027-SafeToOmit

UHXinyang updated a dataset 8 days ago

AgentsSci/ICLR2027-SafeToOmit

View all activity

Organization Card

Community About org cards

AgentsSci — Agents for Science

AgentsSci is a Hugging Face organization for released research code, data, and models.

Resources support agentic evaluation for LLM agents and multi-agent systems: matched protocol traces, reviewer-to-solver critique uptake, answer-repair diagnostics, controlled memory comparisons, and cost-aware protocol routing.

For more research and publications, please see: chihhsuan-yang.github.io.

Precise but Uncoupled — NeurIPS 2026, Main Track
Cost-Aware Protocol Routing — Findings of EMNLP 2026

Manuscripts under review

Wrong but Useful Submitted to AAAI 2027 · Under review · Public preprint Project · Paper · Code · Dataset
Beyond Division of Labor: How Roles Shape Feedback and Recovery in Multi-Agent Systems Submitted to AAMAS 2027 · Under review Project · Code · Dataset
What Is This Memory Useful For? Measuring Task-Conditioned Utility in Long-Term Agent Memory Submitted to AAMAS 2027 · Under review Project · Code · Dataset

Each release identifies its authors, publication status, provenance, license, and version notes. Follow source-dataset usage restrictions; evaluation data is not automatically licensed for model training.

Research contact: Chih-Hsuan (Bella) Yang · bellayang@anl.gov