If your data distribution shifts, use self-learning
University of Tübingen（图宾根大学）；Amazon Tübingen（亚马逊图宾根）；University of Oxford（牛津大学）；Max-Planck Institute for Intelligent Systems Tübingen（马克斯·普朗克智能系统研究所图宾根分所）
Evgenia Rusak, Steffen Schneider, George Pachitariu, Luisa Eck, Peter Gehler, Oliver Bringmann, Wieland Brendel, Matthias Bethge
We demonstrate that self-learning techniques like entropy minimization and pseudo-labeling are simple and effective at improving performance of a deployed computer vision model under systematic domain shifts. We conduct a wide range of large-scale experiments and show consistent improvements irrespective of the model architecture, the pre-training technique or the type of distribution shift. At the same time, self-learning is simple to use in practice because it does not require knowledge or access to the original training data or scheme, is robust to hyperparameter choices, is straight-forward to implement and requires only a few adaptation epochs. This makes self-learning techniques highly attractive for any practitioner who applies machine learning algorithms in the real world. We present state-of-the-art adaptation results on CIFAR10-C (8.5% error), ImageNet-C (22.0% mCE), ImageNet-R (17.4% error) and ImageNet-A (14.8% error), theoretically study the dynamics of self-supervised adaptation methods and propose a new classification dataset (ImageNet-D) which is challenging even with adaptation.
Amazon Web Services（亚马逊网络服务）；Rutgers University（罗格斯大学）
Traditional normalization techniques (e.g., Batch Normalization and Instance Normalization) generally and simplistically assume that training and test data follow the same distribution. As distribution shifts are inevitable in real-world applications, well-trained models with previous normalization methods can perform badly in new environments. Can we develop new normalization methods to improve generalization robustness under distribution shifts? In this paper, we answer the question by proposing CrossNorm and SelfNorm. CrossNorm exchanges channel-wise mean and variance between feature maps t…
University of California, Berkeley（加州大学伯克利分校）；University of California, San Diego（加州大学圣地亚哥分校）
In this paper, we propose Test-Time Training, a general approach for improving the performance of predictive models when training and test data come from different distributions. We turn a single unlabeled test sample into a self-supervised learning problem, on which we update the model parameters before making a prediction. This also extends naturally to data in an online stream. Our simple approach leads to improvements on diverse image classification benchmarks aimed at evaluating robustness to distribution shifts.
Despite their recent success, deep neural networks continue to perform poorly when they encounter distribution shifts at test time. Many recently proposed approaches try to counter this by aligning the model to the new distribution prior to inference. With no labels available this requires unsupervised objectives to adapt the model on the observed test data. In this paper, we propose Test-Time Self-Training (TeST): a technique that takes as input a model trained on some source data and a novel data distribution at test time, and learns invariant and robust representations using a student-teach…
This paper addresses the problem of set-to-set matching, which involves matching two different sets of items based on some criteria, especially in the case of high-dimensional items like images. Although neural networks have been applied to solve this problem, most machine learning-based approaches assume that the training and test data follow the same distribution, which is not always true in real-world scenarios. To address this limitation, we introduce SHIFT15M, a dataset that can be used to evaluate set-to-set matching models when the distribution of data changes between training and testi…
Centre for Research and Technology Hellas（希腊研究与科技中心）；Universidad Politécnica de Madrid（马德里理工大学）
In this work we contribute a distribution shift benchmark for a computer vision task; monocular depth estimation. Our differentiation is the decomposition of the wider distribution shift of uncontrolled testing on in-the-wild data, to three distinct distribution shifts. Specifically, we generate data via synthesis and analyze them to produce covariate (color input), prior (depth output) and concept (their relationship) distribution shifts. We also synthesize combinations and show how each one is indeed a different challenge to address, as stacking them produces increased performance drops and …
Georgia Institute of Technology（佐治亚理工学院）；Samsung Research America（三星美国研究院）
Moving beyond testing on in-distribution data works on Out-of-Distribution (OOD) detection have recently increased in popularity. A recent attempt to categorize OOD data introduces the concept of near and far OOD detection. Specifically, prior works define characteristics of OOD data in terms of detection difficulty. We propose to characterize the spectrum of OOD data using two types of distribution shifts: covariate shift and concept shift, where covariate shift corresponds to change in style, e.g., noise, and concept shift indicates a change in semantics. This characterization reveals that s…
Robustness to distribution shifts is critical for deploying machine learning models in the real world. Despite this necessity, there has been little work in defining the underlying mechanisms that cause these shifts and evaluating the robustness of algorithms across multiple, different distribution shifts. To this end, we introduce a framework that enables fine-grained analysis of various distribution shifts. We provide a holistic analysis of current state-of-the-art methods by evaluating 19 distinct methods grouped into five categories across both synthetic and real-world datasets. Overall, w…
City University of Hong Kong（香港城市大学）；Dalian University of Technology（大连理工大学）；DAMO Academy, Alibaba Group（阿里巴巴集团达摩院）；Tsinghua University（清华大学）
The performance of machine learning models under distribution shift has been the focus of the community in recent years. Most of current methods have been proposed to improve the robustness to distribution shift from the algorithmic perspective, i.e., designing better training algorithms to help the generalization in shifted test distributions. This paper studies the distribution shift problem from the perspective of pre-training and data augmentation, two important factors in the practice of deep learning that have not been systematically investigated by existing work. By evaluating seven pre…
Experiencing domain shifts during test-time is nearly inevitable in practice and likely results in a severe performance degradation. To overcome this issue, test-time adaptation continues to update the initial source model during deployment. A promising direction are methods based on self-training which have been shown to be well suited for gradual domain adaptation, since reliable pseudo-labels can be provided. In this work, we address two problems that exist when applying self-training in the setting of test-time adaptation. First, adapting a model to long test sequences that contain multipl…
We study how robust current ImageNet models are to distribution shifts arising from natural variations in datasets. Most research on robustness focuses on synthetic image perturbations (noise, simulated weather artifacts, adversarial examples, etc.), which leaves open how robustness on synthetic distribution shift relates to distribution shift arising in real data. Informed by an evaluation of 204 ImageNet models in 213 different test conditions, we find that there is often little to no transfer of robustness from current synthetic to natural distribution shift. Moreover, most current techniqu…
University of Texas at Austin（德克萨斯大学奥斯汀分校）；Michigan State University（密歇根州立大学）
Increasing concerns have been raised on deep learning fairness in recent years. Existing fairness-aware machine learning methods mainly focus on the fairness of in-distribution data. However, in real-world applications, it is common to have distribution shift between the training and test data. In this paper, we first show that the fairness achieved by existing methods can be easily broken by slight distribution shifts. To solve this problem, we propose a novel fairness learning method termed CUrvature MAtching (CUMA), which can achieve robust fairness generalizable to unseen domains with unkn…
Pennsylvania State University（宾夕法尼亚州立大学）；Stanford University（斯坦福大学）
Distribution shift occurs when the test distribution differs from the training distribution, and it can considerably degrade performance of machine learning models deployed in the real world. Temporal shifts -- distribution shifts arising from the passage of time -- often occur gradually and have the additional structure of timestamp metadata. By leveraging timestamp metadata, models can potentially learn from trends in past distribution shifts and extrapolate into the future. While recent works have studied distribution shifts, temporal shifts remain underexplored. To address this gap, we cur…
Distribution shifts -- where the training distribution differs from the test distribution -- can substantially degrade the accuracy of machine learning (ML) systems deployed in the wild. Despite their ubiquity in the real-world deployments, these distribution shifts are under-represented in the datasets widely used in the ML community today. To address this gap, we present WILDS, a curated benchmark of 10 datasets reflecting a diverse range of distribution shifts that naturally arise in real-world applications, such as shifts across hospitals for tumor identification; across camera traps for w…
Transfer learning is essential when sufficient data comes from the source domain, with scarce labeled data from the target domain. We develop estimators that achieve minimax linear risk for linear regression problems under distribution shift. Our algorithms cover different transfer learning settings including covariate shift and model shift. We also consider when data are generated from either linear or general nonlinear models. We show that linear minimax estimators are within an absolute constant of the minimax risk even among nonlinear estimators for various source/target distributions.
University of Southampton（南安普顿大学）；Schufa Holding AG（Schufa控股股份公司）；University of Tuebingen（蒂宾根大学）；University of Stuttgart（斯图加特大学）
As input data distributions evolve, the predictive performance of machine learning models tends to deteriorate. In the past, predictive performance was considered the key indicator to monitor. However, explanation aspects have come to attention within the last years. In this work, we investigate how model predictive performance and model explanation characteristics are affected under distribution shifts and how these key indicators are related to each other for tabular data. We find that the modeling of explanation shifts can be a better indicator for the detection of predictive performance ch…
Given an algorithmic predictor that is "fair" on some source distribution, will it still be fair on an unknown target distribution that differs from the source within some bound? In this paper, we study the transferability of statistical group fairness for machine learning predictors (i.e., classifiers or regressors) subject to bounded distribution shifts. Such shifts may be introduced by initial training data uncertainties, user adaptation to a deployed predictor, dynamic environments, or the use of pre-trained models in new settings. Herein, we develop a bound that characterizes such transfe…
Cornell University（康奈尔大学）；Facebook AI Research（Facebook 人工智能研究院）
Machine learning models often encounter distribution shifts when deployed in the real world. In this paper, we focus on adaptation to label distribution shift in the online setting, where the test-time label distribution is continually changing and the model must dynamically adapt to it without observing the true label. Leveraging a novel analysis, we show that the lack of true label does not hinder estimation of the expected test loss, which enables the reduction of online label shift adaptation to conventional online learning. Informed by this observation, we propose adaptation algorithms in…
Rapid progress in representation learning has led to a proliferation of embedding models, and to associated challenges of model selection and practical application. It is non-trivial to assess a model's generalizability to new, candidate datasets and failure to generalize may lead to poor performance on downstream tasks. Distribution shifts are one cause of reduced generalizability, and are often difficult to detect in practice. In this paper, we use the embedding space geometry to propose a non-parametric framework for detecting distribution shifts, and specify two tests. The first test detec…
Machine learning on data streams is increasingly more present in multiple domains. However, there is often data distribution shift that can lead machine learning models to make incorrect decisions. While there are automatic methods to detect when drift is happening, human analysis, often by data scientists, is essential to diagnose the causes of the problem and adjust the system. We propose Data+Shift, a visual analytics tool to support data scientists in the task of investigating the underlying factors of shift in data features in the context of fraud detection. Design requirements were deriv…
A generalized approach to label shift: the Conditional Probability Shift Model
Institute of Computer Science, Polish Academy of Sciences（波兰科学院计算机科学研究所）；Faculty of Mathematics and Information Science, Warsaw University of Technology（华沙理工大学数学与信息科学学院）
University of Maryland, College Park（马里兰大学帕克分校）；Boston University（波士顿大学）
The increasing reliance on ML models in high-stakes tasks has raised a major concern on fairness violations. Although there has been a surge of work that improves algorithmic fairness, most of them are under the assumption of an identical training and test distribution. In many real-world applications, however, such an assumption is often violated as previously trained fair models are often deployed in a different environment, and the fairness of such models has been observed to collapse. In this paper, we study how to transfer model fairness under distribution shifts, a widespread issue in pr…
When a model informs decisions about people, distribution shifts can create undue disparities. However, it is hard for external entities to check for distribution shift, as the model and its training set are often proprietary. In this paper, we introduce and study a black-box auditing method to detect cases of distribution shift that lead to a performance disparity of the model across demographic groups. By extending techniques used in membership and property inference attacks -- which are designed to expose private information from learned models -- we demonstrate that an external auditor can…
University of Konstanz（康斯坦茨大学）；ETH Zurich（苏黎世联邦理工学院）；Ecole Polytechnique Fédérale de Lausanne（洛桑联邦理工学院）
Training models that perform well under distribution shifts is a central challenge in machine learning. In this paper, we introduce a modeling framework where, in addition to training data, we have partial structural knowledge of the shifted test distribution. We employ the principle of minimum discriminating information to embed the available prior knowledge, and use distributionally robust optimization to account for uncertainty due to the limited samples. By leveraging large deviation results, we obtain explicit generalization bounds with respect to the unknown shifted distribution. Lastly,…
We propose a novel approach for semi-supervised learning (SSL) designed to overcome distribution shifts between training and real-world data arising in the keyword spotting (KWS) task. Shifts from training data distribution are a key challenge for real-world KWS tasks: when a new model is deployed on device, the gating of the accepted data undergoes a shift in distribution, making the problem of timely updates via subsequent deployments hard. Despite the shift, we assume that the marginal distributions on labels do not change. We utilize a modified teacher/student training framework, where lab…
Bitdefender（Bitdefender公司）；University Politehnica of Bucharest（布加勒斯特理工大学）；University of Bucharest（布加勒斯特大学）
The human ability to synchronize the feedback from all their senses inspired recent works in multi-task and multi-modal learning. While these works rely on expensive supervision, our multi-task graph requires only pseudo-labels from expert models. Every graph node represents a task, and each edge learns between tasks transformations. Once initialized, the graph learns self-supervised, based on a novel consensus shift algorithm that intelligently exploits the agreement between graph pathways to generate new pseudo-labels for the next learning cycle. We demonstrate significant improvement from o…
As supervised learning still dominates most AI applications, test-time performance is often unexpected. Specifically, a shift of the input covariates, caused by typical nuisances like background-noise, illumination variations or transcription errors, can lead to a significant decrease in prediction accuracy. Recently, it was shown that incorporating self-supervision can significantly improve covariate shift robustness. In this work, we propose Self-Supervised Dynamic Networks (SSDN): an input-dependent mechanism, inspired by dynamic networks, that allows a self-supervised network to predict th…
The Hong Kong Polytechnic University（香港理工大学）；DAMO Academy, Alibaba（阿里巴巴达摩院）
Transformer-based methods have recently achieved great advancement on 2D image-based vision tasks. For 3D video-based tasks such as action recognition, however, directly applying spatiotemporal transformers on video data will bring heavy computation and memory burdens due to the largely increased number of patches and the quadratic complexity of self-attention computation. How to efficiently and effectively model the 3D self-attention of video data has been a great challenge for transformers. In this paper, we propose a Temporal Patch Shift (TPS) method for efficient 3D self-attention modeling…
University of Maryland, Baltimore County（马里兰大学巴尔的摩县分校）；University of California, Davis（加利福尼亚大学戴维斯分校）
Most recent self-supervised learning (SSL) algorithms learn features by contrasting between instances of images or by clustering the images and then contrasting between the image clusters. We introduce a simple mean-shift algorithm that learns representations by grouping images together without contrasting between them or adopting much of prior on the structure of the clusters. We simply "shift" the embedding of each image to be close to the "mean" of its neighbors. Since in our setting, the closest neighbor is always another augmentation of the same image, our model will be identical to BYOL …
Histopathology images are gigapixel-sized and include features and information at different resolutions. Collecting annotations in histopathology requires highly specialized pathologists, making it expensive and time-consuming. Self-training can alleviate annotation constraints by learning from both labeled and unlabeled data, reducing the amount of annotations required from pathologists. We study the design of teacher-student self-training systems for Non-alcoholic Steatohepatitis (NASH) using clinical histopathology datasets with limited annotations. We evaluate the models on in-distribution…
University of Waterloo（滑铁卢大学）；Waterloo Artificial Intelligence Institute（滑铁卢人工智能研究所）
In this study, we propose the Affine Variational Autoencoder (AVAE), a variant of Variational Autoencoder (VAE) designed to improve robustness by overcoming the inability of VAEs to generalize to distributional shifts in the form of affine perturbations. By optimizing an affine transform to maximize ELBO, the proposed AVAE transforms an input to the training distribution without the need to increase model complexity to model the full distribution of affine transforms. In addition, we introduce a training procedure to create an efficient model by learning a subset of the training distribution, …
