EvoSci: A Bio-Inspired Multi-Agent Framework for the Evolution of Scientific Discovery
EvoSci, introduced in a paper on arXiv (2605.24018), is a multi-agent framework for scientific discovery that combines bio-inspired evolution with knowledge graph modeling. It employs role-based agents—mentor, researcher, and reviewer—to iteratively generate, evaluate, and refine research ideas through collaborative reasoning, shared memory, and evolutionary feedback. In experiments on real-world research topics, EvoSci outperformed strong baselines in LLM-based structured peer-review and comparative ranking evaluations, achieving the highest overall peer-review score (ICLR 4.90) and top ranking (Top-10 = 54). These results indicate its superiority in both scientific idea generation and continuous discovery, addressing challenges in research workflow design and multi-role collaboration.
Automates and improves scientific idea generation through multi-agent collaboration.