Reproduce “Combating Catastrophic Forgetting in Continual Domain Adaptation via Knowledge Recasting” 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
Zijie Liu, Zhen Tan, Charles Fleming, Tianlong Chen
SST helps models learn new domains without losing earlier capabilities by adding reusable procedural skill schemas to training examples. It extracts and retrieves schemas from solution traces, while the model predicts the gold response; main evaluation removes retrieved skills. On Qwen2.5-3B-Instruct’s MATH stream, Stream Avg. is 25.12% versus 19.96% for Seq-SFT, and Forget. is 7.10 versus 42.41 points; results cover tested streams, not all settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.