Detecting Overfitting via Adversarial Examples
Roman Werpachowski András György Csaba Szepesvári DeepMind, London, UK
Abstract
The frequent reuse of test sets in popular benchmark problems raises doubts about the credibility of reported test-error rates. Verifying whether a learned model is overfitted to a test set is challenging as independent test sets drawn from the same data distribution are usually unavailable, while other test sets may introduce a distribution shift. We propose a new hypothesis test that uses only the original test data to detect overfitting. It utilizes a new unbiased error estimate that is based on adversarial examples generated from the test data and importance weighting. Overfitting is detected if this error estimate is sufficiently different from the original test error rate. We develop a specialized variant of our test for multiclass image classification, and apply it to testing overfitting of recent models to the popular ImageNet benchmark. Our method correctly indicates overfitting of the trained model to the training set, but is not able to detect any overfitting to the test set, in line with other recent work on this topic.
中文速览
基准测试集（benchmark test set）被全球研究者反复使用，导致模型可能悄悄"记住"了测试集而非真正泛化，但验证这一点极为困难，因为我们通常没有额外的同分布独立测试数据。本文提出一种无需额外数据的统计假设检验方法：利用对抗样本（adversarial examples）对原始测试集进行变换，再通过重要性权重（importance weighting）修正分布偏移，从而构造出一个对"测试集过拟合"更敏感的无偏误差估计量——若该估计值与原始测试误差率相差显著，则判定模型已过拟合测试集。作者将该方法应用于经典图像分类基准 ImageNet，结果显示：检验能正确识别模型对训练集的过拟合，但未能发现任何对 ImageNet 测试集的过拟合迹象，与近期其他研究结论一致。这项工作为评估主流基准测试的可信度提供了一种严谨且无侵入性的量化工具，有助于厘清"刷榜"行为究竟是真实进步还是测试集泄漏的假象。
原文 arXiv:1903.02380;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1903.02380v2