Reproduce “Distributionally Robust Active Learning for Gaussian Process Regression” 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

Shion Takeno, Yoshito Okura, Yu Inatsu, Tatsuya Aoyama, Tomonari Tanaka, Satoshi Akahane, Hiroyuki Hanada, Noriaki Hashimoto and 4 more

Researchers developed a Distributionally Robust Active Learning (DRAL) framework for Gaussian Process Regression (GPR) that minimizes the worst-case expected prediction error over an ambiguity set of possible target distributions. This framework provides the first theoretical guarantees for convergence of the worst-case expected error in GPR, with empirical results demonstrating its superior robustness compared to existing active learning methods on both synthetic and real-world datasets.

Key insights

  • Minimizing the worst-case expected posterior variance of a GPR model is a theoretically sound strategy for reducing its overall worst-case prediction error.
  • Distributionally Robust Active Learning (DRAL) for GPR can achieve theoretical convergence guarantees for its worst-case expected error, driving it to zero with a finite number of samples.
  • Explicitly considering an ambiguity set of target distributions in active learning enhances model robustness against uncertainty or shifts in the true test data distribution.

Summary by alphaXiv.

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