Reproduce “Decoupling Reasoning and Knowledge Injection for In-Context Knowledge Editing” 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
Changyue Wang, Weihang Su, Qingyao Ai, Yujia Zhou, Yiqun Liu
DecKER is an in-context knowledge-editing method for multi-hop questions that plans a sequence of relations before filling masked entities from edited memory. This separates the reasoning route from the facts used along it. On Llama-3.1-8B-Instruct, it reaches 59.0% Multi-Hop Accuracy on MQuAKE-CF-3k-v2, versus 36.0% for EditCoT; results cover tested benchmarks, not every task or model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.