Reproduce “Bayesian Deep Ensembles via the Neural Tangent Kernel” 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
Bobby He, Balaji Lakshminarayanan, Yee Whye Teh
The authors modify deep ensembles by adding an independently sampled, fixed random function to each network during training. In the infinite-width, squared-error setting, this makes the ensemble approximate a Gaussian-process posterior predictive by restoring uncertainty associated with the neural tangent kernel. The method matches the analytic reference in a toy regression example and improves some out-of-distribution results, but not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.