Limitations of Using Identical Distributions for Training and Testing When Learning Boolean Functions

TL;DR

This paper investigates whether training and test data should come from identical distributions for learning Boolean functions, revealing that under certain assumptions, mismatched distributions can sometimes lead to better generalization.

Contribution

It demonstrates that, assuming one-way functions, matching training and test distributions is not always optimal, and explores conditions where standard practices hold.

Findings

Matching distributions is not always optimal under certain cryptographic assumptions.

Optimal learners can benefit from distribution mismatch in some scenarios.

Regularities in target functions can restore the standard conclusion of distribution matching.

Abstract

When the distributions of the training and test data do not coincide, the problem of understanding generalization becomes considerably more complex, prompting a variety of questions. Prior work has shown that, for some fixed learning methods, there are scenarios where training on a distribution different from the test distribution improves generalization. However, these results do not account for the possibility of choosing, for each training distribution, the optimal learning algorithm, leaving open whether the observed benefits stem from the mismatch itself or from suboptimality of the learner. In this work, we address this question in full generality. That is, we study whether it is always optimal for the training distribution to be identical to the test distribution when the learner is allowed to be optimally adapted to the training distribution. Surprisingly, assuming the existence of one-way functions, we find that the answer is no. That is, matching distributions is not always the best scenario. Nonetheless, we also show that when certain regularities are imposed on the target functions, the standard conclusion is recovered in the case of the uniform distribution.