Reproduce “MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents” 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
Haoran Jin, Defu Lian, Xiting Wang, Hong Xie
MUSE helps scientific coding agents use stored experience during implementation. It supplies shared checks and problem-level approaches at task start, then retrieves code-level lessons and gates targeted reviews. One GPU-kernel variance revision passed at 1.884× the benchmark reference. On KernelBenchX, MUSE reached 79.5% correctness and 44.3% acceleration, versus ACE’s 66.4% and 36.9%; evaluation used one agent setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.