arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2203.13834cs.CVcs.AIcs.LG
收藏

一种改进的神经网络训练时间正则化损失

A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved Neural Network Calibration

Indian Institute of Technology Delhi(印度理工学院德里分校);TCS Research(塔塔咨询服务公司研究院)

Ramya Hebbalaguppe, Jatin Prakash, Neelabh Madan, Chetan Arora

摘要:

众所周知,深度神经网络(DNN)会犯过度自信的错误,这使得它们在安全关键型应用中的使用存在问题。最先进的(SOTA)校准技术提高了预测标签的置信度,而非最大类(例如,前2名,前5名)的置信度未校准。这样的校准不适合于使用后处理的标记细化。此外,大多数SOTA技术事后学习一些超参数,从而排除了图像或像素特定校准的范围。这使得它们不适合在域偏移下进行校准,或用于密集预测任务,如语义分割。在本文中,我们主张干预在训练时间本身,以便直接产生校准DNN模型。我们提出了一种新的辅助损失函数:多类置信度和准确度差异(MDCA),以实现相同的MDCA可以与其他应用程序/任务特定的损失函数。我们表明,MDCA训练导致更好的校准模型的预期校准误差(ECE),静态校准误差(SCE)的图像分类,和分割任务。我们在CIFAR 100数据集上报告ECE(SCE)评分为0.72(1.60),而SOTA为1.90(1.71)。在域转移下,使用MDCA在PACS数据集上训练的ResNet-18模型在所有域中的平均ECE(SCE)得分为19.7(9.7),而SOTA为24.2(11.8)。对于分割任务,我们报告了与病灶丢失相比PASCAL-VOC数据集上校准误差减少2倍。最后,MDCA训练改进了校准,即使是在不平衡的数据上,以及自然语言分类任务。我们已经在这里发布了代码:代码可以在https://github.com/mdca-loss上获得

英文摘要:

Deep Neural Networks ( DNN s) are known to make overconfident mistakes, which makes their use problematic in safety-critical applications. State-of-the-art ( SOTA ) calibration techniques improve on the confidence of predicted labels alone and leave the confidence of non-max classes (e.g. top-2, top-5) uncalibrated. Such calibration is not suitable for label refinement using post-processing. Further, most SOTA techniques learn a few hyper-parameters post-hoc, leaving out the scope for image, or pixel specific calibration. This makes them unsuitable for calibration under domain shift, or for dense prediction tasks like semantic segmentation. In this paper, we argue for intervening at the train time itself, so as to directly produce calibrated DNN models. We propose a novel auxiliary loss function: Multi-class Difference in Confidence and Accuracy ( MDCA ), to achieve the same MDCA can be used in conjunction with other application/task-specific loss functions. We show that training with MDCA leads to better-calibrated models in terms of Expected Calibration Error ( ECE ), and Static Calibration Error ( SCE ) on image classification, and segmentation tasks. We report ECE ( SCE ) score of 0.72 (1.60) on the CIFAR 100 dataset, in comparison to 1.90 (1.71) by the SOTA. Under domain shift, a ResNet-18 model trained on PACS dataset using MDCA gives an average ECE ( SCE ) score of 19.7 (9.7) across all domains, compared to 24.2 (11.8) by the SOTA. For the segmentation task, we report a 2X reduction in calibration error on PASCAL - VOC dataset in comparison to Focal Loss. Finally, MDCA training improves calibration even on imbalanced data, and for natural language classification tasks. We have released the code here: code is available at https://github.com/mdca-loss

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑