arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1805.09460stat.MLcs.AIcs.CVcs.LGstat.ME
收藏

谨慎的深度学习

Cautious Deep Learning

Department of Statistics and Data Science, Carnegie Mellon University(卡内基梅隆大学统计与数据科学系);Machine Learning Department, Carnegie Mellon University(卡内基梅隆大学机器学习系)

Yotam Hechtlinger, Barnabás Póczos, Larry Wasserman

摘要:

大多数分类器通过选择条件分布$p(y|x)$的估计的最大值来操作,其中$x$代表要分类的实例的特征,并且$y$表示其标签。这通常会导致{ \em傲慢偏见}:在指定明确标签时过度自信。通常,观察结果集中在一个小体积上,但分类器为整个空间提供明确的预测。我们建议构造包含一组标签而不是单个标签的共形预测集。这些共形预测集包含概率为$1-α$的真标签。我们的构造是基于$p(x|y)$而不是$p(y|x)$,这导致了一个非常谨慎的分类器:当对象与训练示例不相似时,它输出空集合--意思是“我不知道”。我们的方法的一个重要特性是,对抗性攻击很可能被预测为空集,或者也包括真实标签。我们使用从最先进的卷积神经网络获得的高维特征,在ImageNet ILSVRC数据集和CelebA和IMDB-Wiki面部数据集上展示了性能。

英文摘要:

Most classifiers operate by selecting the maximum of an estimate of the conditional distribution $p(y|x)$ where $x$ stands for the features of the instance to be classified and $y$ denotes its label. This often results in a {\em hubristic bias}: overconfidence in the assignment of a definite label. Usually, the observations are concentrated on a small volume but the classifier provides definite predictions for the entire space. We propose constructing conformal prediction sets which contain a set of labels rather than a single label. These conformal prediction sets contain the true label with probability $1-α$. Our construction is based on $p(x|y)$ rather than $p(y|x)$ which results in a classifier that is very cautious: it outputs the null set --- meaning "I don't know" --- when the object does not resemble the training examples. An important property of our approach is that adversarial attacks are likely to be predicted as the null set or would also include the true label. We demonstrate the performance on the ImageNet ILSVRC dataset and the CelebA and IMDB-Wiki facial datasets using high dimensional features obtained from state of the art convolutional neural networks.

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。

↑