Reproduce “Agnostic Learning by Imitating” 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
Sílvia Casacuberta, Parikshit Gopalan, Varun Kanade, Omer Reingold, Konstantinos Stavropoulos, Pranay Tankala
The authors show that adaptive access to a multiaccuracy oracle can yield agnostic learning under expected absolute loss. The reduction combines oracle responses to compete with the best predictor in a specified class, using O(1/ε^5) calls to a (C, Ω(ε²))-multiaccuracy oracle for (C, ε)-agnostic learning. This establishes an oracle-based equivalence, not an empirical result, under the paper’s assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.