Tag
The paper introduces MemIG, a self-evolving memory framework for LLM agents that stores experience as indexable, quantitative, structured records (instead of free-form text) to improve in silico antibody optimization, including epistasis modeling of co-applied mutations.