Reproduce “Online Parameter-Free Learning of Multiple Low Variance Tasks” 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
Giulia Denevi, Dimitris Stamos, Massimiliano Pontil
The authors present an online method that learns a shared bias across sequential tasks without tuning the task-similarity step size. Its aggressive form updates the bias after each example; its lazy form updates at task boundaries. Under bounded inputs and convex Lipschitz losses, bias-estimation overhead scales as O(sqrt(nT)) versus O(n sqrt(T)). In a synthetic setting, aggressive errors fall faster, but unseen-task transfer is not proved for it.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.