Machine Learning in Non-Stationary Environments - Introduction to Covariate Shift Adaptation

By Shai Ben-David, Steffen Bickel, Karsten M. Borgwardt, Michael Brückner, D. Corfield, A. Globerson, Arthur Gretton, L. Hansen, Matthias Hein, Jiayuan Huang and 12 more. 2012

434 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Representation & Self-Supervised Learning

Citations in the corpus

Recommendations as Treatments: Debiasing Learning and Evaluation (2016, methodology): “Weighted ERM approaches have been effective for cost-sensitive classiﬁcation, domain adaptation and co-variate shift (Zadrozny et al., 2003; Bickel et al., 2009; Sugiyama & Kawanabe, 2012).”
Learning explanations that are hard to vary (2020, background): “Covariate Shift [Sugiyama et al., 2007, Quionero-Candela et al., 2009, Sugiyama and Kawanabe, 2012] formalizes the case in which the distribution of the covariates at test time differs from that observed at training.”
Multi-Source Domain Adaptation: A Causal View (2015, background): “DA approaches have many applications in varies areas including natural language processing, computer vision, and biology.”
On causal and anticausal learning (2012, background): “This scenario is referred to as covariate shift (Sugiyama & Kawanabe, 2012).”
Batch Stationary Distribution Estimation (2020, methodology): “One example is learning under covariate shift (Shimodaira, 2000), where the ratio ⌧ can be estimated by different techniques (Gretton et al., 2009; Nguyen et al., 2008; Sugiyama et al., 2008; Sugiyama and Kawanabe, 2012).”
Domain Aggregation Networks for Multi-Source Domain Adaptation (2019, background): “For instance, the covariate shift setting (Shimodaira, 2000; Gretton et al., 2009; Sugiyama & Kawanabe, 2012; Wen et al., 2014) assumes S (cid:54) = T but f S = f T , while concept drift (Jiang & Zhai, 2007; Gama et al., 2014) assumes Finding a domain-invariant feature space by minimizing a…”
Model-Based Policy Gradients with Parameter-Based Exploration by Least-Squares Conditional Density Estimation (2013, methodology): “Let us denote data samples collected with hyper-parameter When the data collecting policy is diﬀerent from the current policy, importance sampling is a useful technique to correct the estimation bias caused by diﬀering distributions [18].”
Robust Learning under Uncertain Test Distributions: Relating Covariate Shift to Model Misspecification (2014, methodology): “To correct the shifted distribution, major efforts have been dedicated to importance reweighing (Quionero-Candela et al., 2009; Sugiyama & Kawanabe, 2012).”
Approximating Likelihood Ratios with Calibrated Discriminative Classifiers (2015, background): “Density ratio estimation is connected to many machine learning fundamental problems, including transfer learning (Sugiyama and Kawanabe, 2012), prob-27 abilistic classiﬁcation and regression (Vapnik, 1998), outlier detection (Hido et al., 2011), and many others.”
Explaining The Efficacy of Counterfactually-Augmented Data (2020, background): “…that while modern machine learning methods perform remarkably well on randomly partitioned holdout data, performance often decays catastrophically under distribution shift (Quionero-Candela et al., 2009; Sugiyama & Kawanabe, 2012; Szegedy et al., 2014; Ovadia et al., 2019; Filos et al., 2020).”
Constructive setting for problems of density ratio estimation (2015, background): “The estimation of the ratio of two probability densities from a given collection of data is a fundamental technique for solving many data mining problems, including data adaptation [1,2], conditional probability and regression estimation [3], estimation of mutual information [4], change detection in”
Combating Reinforcement Learning's Sisyphean Curse with Intrinsic Fear (2016, background): “Outside of reinforcement learning, a number of papers address problems related to non-stationary data, including a book on covariate shift [25].”
Riemannian Approaches in Brain-Computer Interfaces: A Review (2017, neutral): “an estimator of covariance matrices robust to a kind of nonstationarity—namely covariate-shift [77].”
Distribution-Free One-Pass Learning (2017, background): “By directly downweighting the weights of past observations, our approach can deal with general distribution change rather than a specific one, like the concept drift [10] or covariate shift [11], and thuswithout prior distribution change assumptions.”
Understanding the Effect of Out-of-distribution Examples and Interactive Explanations on Human-AI Decision Making (2021, background): “In practice, a critical growing concern is distribution shift [19, 55, 58].”