Reproduce “Robust Meta-learning for Mixed Linear Regression with Small Batches” 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
Weihao Kong, Raghav Somani, Sham Kakade, Sewoong Oh
The authors study meta-learning for tasks drawn from a mixture of k linear regressions with corrupted tasks. They estimate a shared subspace from small tasks and use higher-order statistics and filtering to cluster larger tasks. Under stated conditions, the heavy-task batch requirement falls from order square root of k to order log k, at the cost of many more tasks and computation; this is a theoretical guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.