EvoScientist: Multi-Agent Evolution for End-to-End Scientific Discovery скачать в хорошем качестве

EvoScientist: Multi-Agent Evolution for End-to-End Scientific Discovery

EvoScientist is a self-evolving multi-agent framework that uses a large-scale language model (LLM) to automate the entire process of scientific discovery. The system consists of a researcher agent who generates research ideas, an engineer agent who implements and executes experimental code, and an evolutionary management agent who analyzes past successes and failures and transforms them into knowledge. Unlike existing systems that rely on a fixed pipeline, continuous memory accumulates promising research directions and effective experimental strategies to improve performance on its own. As a result of the experiment, this framework has shown a performance that overwhelms existing major systems in terms of novelty and validity of ideas. As a result, EvoScientist demonstrated his ability to write an academic journal-level paper independently by reducing repetitive trial and error and increasing the success rate of code execution. https://arxiv.org/pdf/2603.08127