Image classifiers often use spurious patterns, such as "relying on the presence of a person to detect a tennis racket, which do not generalize. In this work, we present an end-to-end pipeline for identifying and mitigating spurious patterns for such models, under the assumption that we have access to pixel-wise object-annotations. We start by identifying patterns such as "the model's prediction for tennis racket changes 63% of the time if we hide the people." Then, if a pattern is spurious, we mitigate it via a novel form of data augmentation. We demonstrate that our method identifies a diverse set of spurious patterns and that it mitigates them by producing a model that is both more accurate on a distribution where the spurious pattern is not helpful and more robust to distribution shift.
Massachusetts Institute of Technology(麻省理工学院);Stanford University(斯坦福大学);Google AI(谷歌人工智能)
我们调查是否事后模型的解释是有效的诊断模型错误-模型调试。为了应对解释模型预测的挑战,已经提出了大量的解释方法。尽管使用越来越多,但尚不清楚它们是否有效。首先,我们根据其来源将\textit{bugs}分类为:~ \textit{data, model, and test-time}污染缺陷。对于几种解释方法,我们评估它们的能力:检测虚假的相关性伪影(数据污染),诊断错误标记的训练样本(数据污染),区分(部分)重新初始化的模型和训练的模型(模型污染),以及检测分布外的输入(测试时污染)。我们发现,测试的方法能够诊断一个虚假的背景错误,但不能最终确定错误标记的训练样本。此外,一类修改反向传播算法的方法对于深度网络的高层参数是不变的;因此,对于诊断模型污染是无效的。我们补充我们的分析与人类受试者的研究,发现受试者无法识别有缺陷的模型使用属性,而是主要依赖于模型预测。总之,我们的研究结果提供了指导的从业者和研究人员转向解释模型调试的工具。
The Hong Kong University of Science and Technology(香港科技大学);Tsinghua University(清华大学);Fudan University(复旦大学);University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)
Towards Musically Meaningful Explanations Using Source Separation
Institute of Computational Perception, Johannes Kepler University(约翰内斯·开普勒大学计算感知研究所);LIT Artificial Intelligence Lab, Johannes Kepler University(约翰内斯·开普勒大学LIT人工智能实验室);Northwestern University(西北大学)
Finding and Removing Clever Hans: Using Explanation Methods to Debug and Improve Deep Models
Technische Universität Berlin(柏林工业大学);Fraunhofer Heinrich Hertz Institute(弗劳恩霍夫海因里希·赫兹研究所);Korea University(高丽大学);Max Planck Institut für Informatik(马克斯·普朗克信息学研究所)