Generating confidence calibrated outputs is of utmost importance for the applications of deep neural networks in safety-critical decision-making systems. The output of a neural network is a probability distribution where the scores are estimated confidences of the input belonging to the corresponding classes, and hence they represent a complete estimate of the output likelihood relative to all classes. In this paper, we propose a novel form of label smoothing to improve confidence calibration. Since different classes are of different intrinsic similarities, more similar classes should result in closer probability values in the final output. This motivates the development of a new smooth label where the label values are based on similarities with the reference class. We adopt different similarity measurements, including those that capture feature-based similarities or semantic similarity. We demonstrate through extensive experiments, on various datasets and network architectures, that our approach consistently outperforms state-of-the-art calibration techniques including uniform label smoothing.
View on arXiv