Understanding and Testing Generalization of Deep Networks on Out-of-Distribution Data
This addresses the critical issue of unreliable model evaluation in real-world scenarios for AI practitioners, but it is incremental as it builds on existing OOD generalization research.
The study tackled the problem that deep networks perform well on in-distribution data but fail on out-of-distribution data, by analyzing experimental test failures and proposing new OOD test paradigms to evaluate generalization, with observations showing ID tests fail to reflect actual performance under distribution shifts.
Deep network models perform excellently on In-Distribution (ID) data, but can significantly fail on Out-Of-Distribution (OOD) data. While developing methods focus on improving OOD generalization, few attention has been paid to evaluating the capability of models to handle OOD data. This study is devoted to analyzing the problem of experimental ID test and designing OOD test paradigm to accurately evaluate the practical performance. Our analysis is based on an introduced categorization of three types of distribution shifts to generate OOD data. Main observations include: (1) ID test fails in neither reflecting the actual performance of a single model nor comparing between different models under OOD data. (2) The ID test failure can be ascribed to the learned marginal and conditional spurious correlations resulted from the corresponding distribution shifts. Based on this, we propose novel OOD test paradigms to evaluate the generalization capacity of models to unseen data, and discuss how to use OOD test results to find bugs of models to guide model debugging.