Reproduce “Implicit Assumptions About Tasks in Continual 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
Kathrin Korte, Christian Metzner, Akshay Bedhotiya, Viet Anh Khoa Tran, Eleni Nisioti, Sebastian Risi
The authors examine how task boundaries affect continual-learning evaluation, distinguishing the experience stream, its task partition, and the learner. In a controlled teacher–student experiment, they hold the trajectory and total training budget fixed while varying partition granularity, and separately vary task order. Both interventions change final test-error profiles, showing sensitivity in this toy setup, not establishing effects on standard benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.