Neural networks often learn to make predictions that overly rely on spurious correlation existing in the dataset, which causes the model to be biased. While previous work tackles this issue by using explicit labeling on the spuriously correlated attributes or presuming a particular bias type, we instead utilize a cheaper, yet generic form of human knowledge, which can be widely applicable to various types of bias. We first observe that neural networks learn to rely on the spurious correlation only when it is "easier" to learn than the desired knowledge, and such reliance is most prominent during the early phase of training. Based on the observations, we propose a failure-based debiasing scheme by training a pair of neural networks simultaneously. Our main idea is twofold; (a) we intentionally train the first network to be biased by repeatedly amplifying its "prejudice", and (b) we debias the training of the second network by focusing on samples that go against the prejudice of the biased network in (a). Extensive experiments demonstrate that our method significantly improves the training of the network against various types of biases in both synthetic and real-world datasets. Surprisingly, our framework even occasionally outperforms the debiasing methods requiring explicit supervision of the spuriously correlated attributes.
A Statistical Framework for Alignment with Biased AI Feedback
Center for Data Science, Zhejiang University（浙江大学数据科学中心）；Department of Electrical and Computer Engineering, Rutgers University（罗格斯大学电气与计算机工程系）；Department of Statistics, Rutgers University（罗格斯大学统计学系）；Faculty of Business and Economics, The University of Hong Kong（香港大学商学院）
Seoul National University（首尔国立大学）；University of Tübingen（图宾根大学）；Idiap Research Institute（Idiap研究所）；NAVER AI Lab（NAVER人工智能实验室）
Neural networks trained with ERM (empirical risk minimization) sometimes learn unintended decision rules, in particular when their training data is biased, i.e., when training labels are strongly correlated with undesirable features. To prevent a network from learning such features, recent methods augment training data such that examples displaying spurious correlations (i.e., bias-aligned examples) become a minority, whereas the other, bias-conflicting examples become prevalent. However, these approaches are sometimes difficult to train and scale to real-world data because they rely on genera…
Learning Decomposable and Debiased Representations via Attribute-Centric Information Bottlenecks
School of Computing and Augmented Intelligence（计算与增强智能学院）；Arizona State University（亚利桑那州立大学）；School of Arts, Media and Engineering（艺术、媒体与工程学院）；School of Life Sciences（生命科学学院）
提出去偏框架 Debiasing Global Workspace，利用基于注意力的信息瓶颈学习可分解的内在与偏置属性组合表示，提升模型分布外泛化能力并区分内在与偏置特征。
Pseudo-labels are confident predictions made on unlabeled target data by a classifier trained on labeled source data. They are widely used for adapting a model to unlabeled data, e.g., in a semi-supervised learning setting. Our key insight is that pseudo-labels are naturally imbalanced due to intrinsic data similarity, even when a model is trained on balanced source data and evaluated on balanced target data. If we address this previously unknown imbalanced classification problem arising from pseudo-labels instead of ground-truth training labels, we could remove model biases towards false majo…
Balancing Unobserved Confounding with a Few Unbiased Ratings in Debiased Recommendations
Peking University（北京大学）；University of Chinese Academy of Sciences（中国科学院大学）；University of California, San Diego（加州大学圣地亚哥分校）；Beijing Technology and Business University（北京工商大学）
Image classification models tend to make decisions based on peripheral attributes of data items that have strong correlation with a target variable (i.e., dataset bias). These biased models suffer from the poor generalization capability when evaluated on unbiased datasets. Existing approaches for debiasing often identify and emphasize those samples with no such correlation (i.e., bias-conflicting) without defining the bias type in advance. However, such bias-conflicting samples are significantly scarce in biased datasets, limiting the debiasing capability of these approaches. This paper first …
Beijing University of Posts and Telecommunications（北京邮电大学）；Peng Cheng Laboratory（鹏城实验室）；Zhejiang University（浙江大学）
Most existing Graph Neural Networks (GNNs) are proposed without considering the selection bias in data, i.e., the inconsistent distribution between the training set with test set. In reality, the test data is not even available during the training process, making selection bias agnostic. Training GNNs with biased selected nodes leads to significant parameter estimation bias and greatly impacts the generalization ability on test nodes. In this paper, we first present an experimental investigation, which clearly shows that the selection bias drastically hinders the generalization ability of GNNs…
With the deluge of digitized information in the Big Data era, massive datasets are becoming increasingly available for learning predictive models. However, in many practical situations, the poor control of the data acquisition processes may naturally jeopardize the outputs of machine learning algorithms, and selection bias issues are now the subject of much attention in the literature. The present article investigates how to extend Empirical Risk Minimization, the principal paradigm in statistical learning, when training observations are generated from biased models, i.e., from distributions t…
Machine learning systems are often trained using data collected from historical decisions. If past decisions were biased, then automated systems that learn from historical data will also be biased. We propose a black-box approach to identify and remove biased training data. Machine learning models trained on such debiased data (a subset of the original training data) have low individual discrimination, often 0%. These models also have greater accuracy and lower statistical disparity than models trained on the full historical data. We evaluated our methodology in experiments using 6 real-world …
Arizona State University（亚利桑那州立大学）；City University of Hong Kong（香港城市大学）
Recommender systems aim to recommend new items to users by learning user and item representations. In practice, these representations are highly entangled as they consist of information about multiple factors, including user's interests, item attributes along with confounding factors such as user conformity, and item popularity. Considering these entangled representations for inferring user preference may lead to biased recommendations (e.g., when the recommender model recommends popular items even if they do not align with the user's interests). Recent research proposes to debias by modeling …
It is widely known in the machine learning community that class noise can be (and often is) detrimental to inducing a model of the data. Many current approaches use a single, often biased, measurement to determine if an instance is noisy. A biased measure may work well on certain data sets, but it can also be less effective on a broader set of data sets. In this paper, we present noise identification using classifier diversity (NICD) -- a method for deriving a less biased noise measurement and integrating it into the learning process. To lessen the bias of the noise measure, NICD selects a div…
Deep neural networks achieve remarkable performances on a wide range of tasks with the aid of large-scale labeled datasets. Yet these datasets are time-consuming and labor-exhaustive to obtain on realistic tasks. To mitigate the requirement for labeled data, self-training is widely used in semi-supervised learning by iteratively assigning pseudo labels to unlabeled samples. Despite its popularity, self-training is well-believed to be unreliable and often leads to training instability. Our experimental studies further reveal that the bias in semi-supervised learning arises from both the problem…
Most NLP datasets are not annotated with protected attributes such as gender, making it difficult to measure classification bias using standard measures of fairness (e.g., equal opportunity). However, manually annotating a large dataset with a protected attribute is slow and expensive. Instead of annotating all the examples, can we annotate a subset of them and use that sample to estimate the bias? While it is possible to do so, the smaller this annotated sample is, the less certain we are that the estimate is close to the true bias. In this work, we propose using Bernstein bounds to represent…
University of California, Irvine（加州大学欧文分校）；Harvard University（哈佛大学）
As machine learning black boxes are increasingly being deployed in domains such as healthcare and criminal justice, there is growing emphasis on building tools and techniques for explaining these black boxes in an interpretable manner. Such explanations are being leveraged by domain experts to diagnose systematic errors and underlying biases of black boxes. In this paper, we demonstrate that post hoc explanations techniques that rely on input perturbations, such as LIME and SHAP, are not reliable. Specifically, we propose a novel scaffolding technique that effectively hides the biases of any g…
Technion – Israel Institute of Technology（以色列理工学院）
We consider the problem of uncertainty estimation in the context of (non-Bayesian) deep neural classification. In this context, all known methods are based on extracting uncertainty signals from a trained network optimized to solve the classification problem at hand. We demonstrate that such techniques tend to introduce biased estimates for instances whose predictions are supposed to be highly confident. We argue that this deficiency is an artifact of the dynamics of training with SGD-like optimizers, and it has some properties similar to overfitting. Based on this observation, we develop an u…
Shortcut Learning in Binary Classifier Black Boxes: Applications to Voice Anti-Spoofing and Biometrics
TCG CREST；Carnegie Mellon University（卡内基梅隆大学）；School of Computing, University of Eastern Finland（东部芬兰大学计算机学院）；University of Eastern Finland（东部芬兰大学）；University of Oulu（奥卢大学）
Federated Learning (FL) has become a popular distributed learning paradigm that involves multiple clients training a global model collaboratively in a data privacy-preserving manner. However, the data samples usually follow a long-tailed distribution in the real world, and FL on the decentralized and long-tailed data yields a poorly-behaved global model severely biased to the head classes with the majority of the training samples. To alleviate this issue, decoupled training has recently been introduced to FL, considering it has achieved promising results in centralized long-tailed learning by …
Causality and Independence Enhancement for Biased Node Classification
Data Intelligence System Research Center, Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所数据智能系统研究中心）；University of Chinese Academy of Sciences（中国科学院大学）；School of Cyber Science and Technology, University of Science and Technology of China（中国科学技术大学网络空间安全学院）；National Engineering Research Center for Public Safety Risk Perception and Control by Big Data (RPP), China Academic of Electronics and Information Technology（中国电子科技集团公司大数据公共安全风险感知与控制国家工程研究中心）
Australian Centre for Visual Technologies（澳大利亚视觉技术中心）；School of Computer Science, The University of Adelaide（阿德莱德大学计算机科学学院）
Cascade classifiers are widely used in real-time object detection. Different from conventional classifiers that are designed for a low overall classification error rate, a classifier in each node of the cascade is required to achieve an extremely high detection rate and moderate false positive rate. Although there are a few reported methods addressing this requirement in the context of object detection, there is no principled feature selection method that explicitly takes into account this asymmetric node learning objective. We provide such an algorithm here. We show that a special case of the…
IBM Research（IBM研究院）；MIT-IBM Watson AI Lab（麻省理工-IBM沃森人工智能实验室）
In consequential decision-making applications, mitigating unwanted biases in machine learning models that yield systematic disadvantage to members of groups delineated by sensitive attributes such as race and gender is one key intervention to strive for equity. Focusing on demographic parity and equality of opportunity, in this paper we propose an algorithm that improves the fairness of a pre-trained classifier by simply dropping carefully selected training data points. We select instances based on their influence on the fairness metric of interest, computed using an infinitesimal jackknife-ba…
Université de Lorraine（洛林大学）；CNRS（法国国家科学研究中心）；Inria（法国国家信息与自动化研究所）；LORIA（洛林研究与信息学实验室（此处按机构含义直译，若有更常见中文名需调整，但根据要求，LORIA作为机构名本身，保留英文并翻译））
Algorithmic decisions are now being used on a daily basis, and based on Machine Learning (ML) processes that may be complex and biased. This raises several concerns given the critical impact that biased decisions may have on individuals or on society as a whole. Not only unfair outcomes affect human rights, they also undermine public trust in ML and AI. In this paper we address fairness issues of ML models based on decision outcomes, and we show how the simple idea of "feature dropout" followed by an "ensemble approach" can improve model fairness. To illustrate, we will revisit the case of "Li…
Data heterogeneity is an inherent challenge that hinders the performance of federated learning (FL). Recent studies have identified the biased classifiers of local models as the key bottleneck. Previous attempts have used classifier calibration after FL training, but this approach falls short in improving the poor feature representations caused by training-time classifier biases. Resolving the classifier bias dilemma in FL requires a full understanding of the mechanisms behind the classifier. Recent advances in neural collapse have shown that the classifiers and feature prototypes under perfec…
Federated learning (FL) provides a privacy-preserving solution for distributed machine learning tasks. One challenging problem that severely damages the performance of FL models is the co-occurrence of data heterogeneity and long-tail distribution, which frequently appears in real FL applications. In this paper, we reveal an intriguing fact that the biased classifier is the primary factor leading to the poor performance of the global model. Motivated by the above finding, we propose a novel and privacy-preserving FL method for heterogeneous and long-tailed data via Classifier Re-training with …
Sharif University of Technology（谢里夫理工大学）；Technical University of Munich（慕尼黑工业大学）；Johns Hopkins University（约翰斯·霍普金斯大学）
Disease prediction is a well-known classification problem in medical applications. GCNs provide a powerful tool for analyzing the patients' features relative to each other. This can be achieved by modeling the problem as a graph node classification task, where each node is a patient. Due to the nature of such medical datasets, class imbalance is a prevalent issue in the field of disease prediction, where the distribution of classes is skewed. When the class imbalance is present in the data, the existing graph-based classifiers tend to be biased towards the major class(es) and neglect the sampl…
A deep neural network (DNN) consists of a nonlinear transformation from an input to a feature representation, followed by a common softmax linear classifier. Though many efforts have been devoted to designing a proper architecture for nonlinear transformation, little investigation has been done on the classifier part. In this paper, we show that a properly designed classifier can improve robustness to adversarial attacks and lead to better prediction results. Specifically, we define a Max-Mahalanobis distribution (MMD) and theoretically show that if the input distributes as a MMD, the linear d…
We prove that the binary classifiers of bit strings generated by random wide deep neural networks with ReLU activation function are biased towards simple functions. The simplicity is captured by the following two properties. For any given input bit string, the average Hamming distance of the closest input bit string with a different classification is at least sqrt(n / (2π log n)), where n is the length of the string. Moreover, if the bits of the initial string are flipped randomly, the average number of flips required to change the classification grows linearly with n. These results are confir…
Yale University（耶鲁大学）；University of Texas at Austin（德克萨斯大学奥斯汀分校）；Amazon AWS AI（亚马逊云服务人工智能）；Fiddler AI
In hybrid human-machine deferral frameworks, a classifier can defer uncertain cases to human decision-makers (who are often themselves fallible). Prior work on simultaneous training of such classifier and deferral models has typically assumed access to an oracle during training to obtain true class labels for training samples, but in practice there often is no such oracle. In contrast, we consider a "closed" decision-making pipeline in which the same fallible human decision-makers used in deferral also provide training labels. How can imperfect and biased human expert labels be used to train a…
Background: Neural networks produce biased classification results due to correlation bias (they learn correlations between their inputs and outputs to classify samples, even when those correlations do not represent cause-and-effect relationships). Objective: This study introduces a fully unsupervised method of mitigating correlation bias, demonstrated with sentiment classification on COVID-19 social media data. Methods: Correlation bias in sentiment classification often arises in conversations about controversial topics. Therefore, this study uses adversarial learning to contrast clusters base…
