Reproduce “Lifelong Context Adaptation through Bidirectional Memory-Skill Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tianyi Xu, Huazheng Wang
MemCryst links episodic evidence and reusable skills through a loop in which task outcomes update retrieval influence for served records, while a frozen language model uses facts and skill text together. On a 600-task ALFWorld ablation, joint serving achieved 63.17% success, versus 59.83% for memory-only and 50.50% for skill-only. These results support complementarity in this setup, not universal benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.