Reproduce “Maximum Likelihood Estimation for Learning Populations of Parameters” 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

Ramya Korlakai Vinayak, Weihao Kong, Gregory Valiant, Sham M. Kakade

The maximum-likelihood estimator infers the population distribution of people’s Bernoulli probabilities from pooled success counts, rather than treating observed fractions as true probabilities. For t=O(log N), its Wasserstein-1 error is Oδ(1/t); for t from Ω(log N) to O(N^{2/9−ε}), it is Oδ(1/√(t log N)). Both match minimax rates in these regimes; these are theoretical guarantees, not empirical results.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.