Reproduce “Knowledge Interaction in Hypernetwork-based Continual Learning: Conflict and Composition” 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
Jiaqi Zheng, Yi Hu, Yadong Mu, Muhan Zhang
SHINE generates a LoRA adapter from context while keeping the base model frozen, adding another route for knowledge without ensuring reliable use. In TRCBENCH, the model followed the contextual answer in 57.10% of examples, versus 27.20% for the adapter answer and 8.75% for the base-model answer. Further tests found unreliable fact composition involving LoRA in this setup, not across all systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.