Mitigating Dataset Bias by Using Per-sample Gradient

Abstract

The performance of deep neural networks is strongly influenced by the training dataset setup. In particular, when attributes having a strong correlation with the target attribute are present, the trained model can provide unintended prejudgments and show significant inference errors (i.e., the dataset bias problem). Various methods have been proposed to mitigate dataset bias, and their emphasis is on weakly correlated samples, called bias-conflicting samples. These methods are based on explicit bias labels involving human or empirical correlation metrics (e.g., training loss). However, such metrics require human costs or have insufficient theoretical explanation. In this study, we propose a debiasing algorithm, called PGD (Per-sample Gradient-based Debiasing), that comprises three steps: (1) training a model on uniform batch sampling, (2) setting the importance of each sample in proportion to the norm of the sample gradient, and (3) training the model using importance-batch sampling, whose probability is obtained in step (2). Compared with existing baselines for various synthetic and real-world datasets, the proposed method showed state-of-the-art accuracy for a the classification task. Furthermore, we describe theoretical understandings about how PGD can mitigate dataset bias. IntroductionDeep neural networks (DNNs) have been a research focus over the past decade owing to their humanlike performance in a range of tasks, including image classification [37], object detection [17], and image generation [18]. Despite their promising performance, the difficulties in obtaining wellcurated training datasets hinder the use of DNNs in real-world applications. In particular, dataset bias [63, 56], a bad training dataset problem, occurs when non-predictive attributes having a high correlation with the target attribute are present. This unintentionally correlated information (i.e., bias attributes) causes prejudice, and thereby leads to inaccurate inferences. For example, most "camel" images include a "desert background," and this unintended correlation can provide a false shortcut for answering "camel" on the basis of the the "desert." In [50,40], samples of data that have a strong correlation (like "desert background" in "camel" class images) are called "bias-aligned samples," while samples of data that have a weak correlation (like "camel on the grass" images) are termed "bias-conflicting samples."To reduce the dataset bias, initial studies [33,47,57,44] frequently assumed a case where labels with bias attributes are provided, but these additional labels provided through human effort are expensive. Alternatively, the bias-type, such as "background" ws assumed in [41,16,5,7,11]. However, assuming biased knowledge from humans is still unreasonable, since even humans cannot predict the type of bias that may exist in a large dataset [54]. Data for deep learning is typically collected by web-crawling without thorough consideration of the dataset bias problem.Recent studies [38,50,34,40] have replaced human intervention with D...