Reproduce “Training independent subnetworks for robust prediction” 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

Marton Havasi, Rodolphe Jenatton, Stanislav Fort, Jeremiah Zhe Liu, Jasper Snoek, Balaji Lakshminarayanan, Andrew M. Dai, Dustin Tran

MIMO turns a network into an ensemble by training output heads on independently sampled inputs, each matched to its label, then repeating one input across slots and averaging predictions in one forward pass. On ResNet28-10/CIFAR100, M=3 raises clean accuracy from 79.8% to 82.0% and corrupted accuracy from 51.4% to 53.7%, but trails a four-member deep ensemble on accuracy and NLL.

Summary by alphaXiv.

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