Reproduce “Learning Neural Networks with Distribution Shift: Efficiently Certifiable Guarantees” 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
Gautam Chandrasekaran, Adam R. Klivans, Lin Lin Lee, Konstantinos Stavropoulos
The authors develop a certify-or-reject method for regression under distribution shift. Using labeled training data and unlabeled test features, it builds a compact representation and tests whether prediction differences transfer. For one-hidden-layer sigmoid networks with hypercontractive training features and bounded labels, accepted predictors have test RMSE at most opt + λ + ε, with failure probability 0.1; test features are unrestricted.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.