The Limitations of Model Uncertainty in Adversarial Settings

Authors: Kathrin Grosse, David Pfaff, Michael Thomas Smith, Michael Backes
Published: 2018-12-06 | Updated: 2019-11-17

Source: https://arxiv.org/abs/1812.02606

PDF: https://arxiv.org/pdf/1812.02606

AIにより推定されたラベル

敵対的サンプル ロバスト性評価 モデルの頑健性保証

※ こちらのラベルはAIによって自動的に追加されました。そのため、正確でないことがあります。
詳細は文献データベースについてをご覧ください。

Abstract

Machine learning models are vulnerable to adversarial examples: minor perturbations to input samples intended to deliberately cause misclassification. While an obvious security threat, adversarial examples yield as well insights about the applied model itself. We investigate adversarial examples in the context of Bayesian neural network’s (BNN’s) uncertainty measures. As these measures are highly non-smooth, we use a smooth Gaussian process classifier (GPC) as substitute. We show that both confidence and uncertainty can be unsuspicious even if the output is wrong. Intriguingly, we find subtle differences in the features influencing uncertainty and confidence for most tasks.