Reliable confidence estimation for the predictions is important in many safety-critical applications. However, modern deep neural networks are often overconfident for their incorrect predictions. Recently, many calibration methods have been proposed to alleviate the overconfidence problem. With calibrated confidence, a primary and practical purpose is to detect misclassification errors by filtering out low-confidence predictions (known as failure prediction). In this paper, we find a general, widely-existed but actually-neglected phenomenon that most confidence calibration methods are useless or harmful for failure prediction. We investigate this problem and reveal that popular confidence calibration methods often lead to worse confidence separation between correct and incorrect samples, making it more difficult to decide whether to trust a prediction or not. Finally, inspired by the natural connection between flat minima and confidence separation, we propose a simple hypothesis: flat minima is beneficial for failure prediction. We verify this hypothesis via extensive experiments and further boost the performance by combining two different flat minima techniques. Our code is available at https://github.com/Impression2805/FMFP

翻译：在众多安全关键型应用中，对预测结果进行可靠的置信度估计至关重要。然而，现代深度神经网络往往对其错误预测过于自信。近年来，为缓解过度自信问题，研究者提出了许多校准方法。通过校准后的置信度，一个主要且实际的应用是过滤掉低置信度的预测以检测误分类错误（即失败预测）。本文发现一个普遍存在但实际被忽视的现象：大多数置信度校准方法对失败预测无效甚至有害。我们研究这一问题并揭示，流行的置信度校准方法通常会导致正确样本与错误样本之间的置信度区分度变差，从而使得决定是否信任某个预测变得更加困难。最后，受平坦最小值与置信度区分度之间自然联系的启发，我们提出一个简单假设：平坦最小值有利于失败预测。我们通过大量实验验证了这一假设，并进一步结合两种不同的平坦最小值技术提升了性能。我们的代码可在 https://github.com/Impression2805/FMFP 获取。