Reproduce “A Generalization Theory based on Independent and Task-Identically Distributed Assumption” 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

Guanhua Zheng, Jitao Sang, Houqiang Li, Jian Yu, Changsheng Xu

The authors introduce ITID, which allows independent examples to differ in task-uncorrelated factors while requiring task-correlated factors and labels to share a marginal distribution. Its bounds penalize prediction dependence on irrelevant factors and motivate balancing them. On CIFAR-10, peripheral random erasing lowered test error across three CNNs, while center-biased erasing performed worse than uniform; this is not a universal rule.

Summary by alphaXiv.

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