NanoResearch: Co-Evolving Skills, Memory, and Policy for Personalized Research Automation
Researchers have introduced NanoResearch, a novel multi-agent framework designed to personalize research automation powered by Large Language Models (LLMs). While current AI systems can automate the research pipeline from ideation to writing, they often produce uniform outputs that fail to account for individual researchers' unique resource constraints, methodological preferences, and target formats. NanoResearch addresses this limitation through a tri-level co-evolution mechanism. It features a skill bank that distills recurring operations into reusable procedural rules, a memory module that retains user-specific and project-specific experiences to ground planning decisions, and a label-free policy learning system that converts free-form feedback into persistent planner updates. These three components co-evolve, where reliable skills enhance memory, richer memory improves planning, and preference internalization continuously realigns the system to the user. Extensive experiments indicate that NanoResearch significantly outperforms state-of-the-art AI research systems, progressively refining its output quality while reducing costs over successive cycles, thereby making research automation genuinely usable for diverse individual needs.
Wire timeline
NanoResearch: Co-Evolving Skills, Memory, and Policy for Personalized Research Automation
Researchers have introduced NanoResearch, a novel multi-agent framework designed to personalize research automation powered by Large Language Models (LLMs). While current AI systems can automate the research pipeline from ideation to writing, they often produce uniform outputs that fail to account for individual researchers' unique resource constraints, methodological preferences, and target formats. NanoResearch addresses this limitation through a tri-level co-evolution mechanism. It features a skill bank that distills recurring operations into reusable procedural rules, a memory module that retains user-specific and project-specific experiences to ground planning decisions, and a label-free policy learning system that converts free-form feedback into persistent planner updates. These three components co-evolve, where reliable skills enhance memory, richer memory improves planning, and preference internalization continuously realigns the system to the user. Extensive experiments indicate that NanoResearch significantly outperforms state-of-the-art AI research systems, progressively refining its output quality while reducing costs over successive cycles, thereby making research automation genuinely usable for diverse individual needs.
cs.AI updates on arXiv.org