Discriminative learning for differing training and test distributions

By Steffen Bickel, Michael Brückner, Tobias Scheffer. 2007

486 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Representation & Self-Supervised Learning

Citations in the corpus

A theory of learning from different domains (2010, methodology): “This line of work leads to algorithms based on instance weighting, which have also been explored empirically in the machine learning and natural language processing communities (Jiang and Zhai 2007; Bickel et al. 2007).”
GenDICE: Generalized Offline Estimation of Stationary Values (2020, background): “Classical density ratio estimation includes moment matching (Gretton et al., 2008), probabilistic classification (Bickel et al., 2007), and ratio matching (Nguyen et al., 2008; Sugiyama et al., 2008; Kanamori et al., 2009).”
DualDICE: Behavior-Agnostic Estimation of Discounted Stationary Distribution Corrections (2019, methodology): “error), various direct ratio estimators have been proposed [37], including the moment matching approach [13], probabilistic classification approach [3, 5, 33], and ratio matching approach [17, 27, 38] The proposed DualDICE algorithm, as a direct approach for density ratio estimation, bears some simi”
Bayesian Multitask Learning with Latent Hierarchies (2009, background): “Similar ideas have been exploited in discriminative models for domain adaptation [2].”
A Review of Domain Adaptation without Target Labels (2019, methodology): “Lastly, it is also possible to avoid the two-step procedure, and optimize the weights simultaneously with optimizing the classifier [110], [111].”
A unifying view on dataset shift in classification (2012, background): “‘‘Differing training and test distributions’’ [8], who define it as follows (the two definitions appear in different places in the same paper): – ‘‘The training instances are governed by a distribution that is allowed to differ arbitrarily from the test distribution.”
Gesture Recognition Portfolios for Personalization (2014, methodology): “Such an approach to personalization is intuitive and is analogous to the instance weighting approaches used in natural language parsing [1, 10, 5].”
DORO: Distributional and Outlier Robust Optimization (2021, unknown): “Related Work Distributional shift naturally arises in many machine learning applications and has been widely studied in statistics, applied probability and optimization ( Shimodaira, 2000; Huang et al., 2006; Bickel et al., 2007; Quionero-Candela et al., 2009).”
Conditional Density Estimation via Least-Squares Density Ratio Estimation (2010, methodology): “The kernel mean matching method (Huang et al., 2007) and the logistic regression based method (Qin, 1998; Cheng & Chu, 2004; Bickel et al., 2007) also allow one to directly estimate a density ratio q(x)/q′(x).”
Information-Theoretical Learning of Discriminative Clusters for Unsupervised Domain Adaptation (2012, background): “For instance, in covariate shift (Shimodaira, 2000; Bickel et al., 2007; Huang et al., 2007), the marginal distributions of the features are diﬀerent across domains while the posterior distribution of the label remains the same.”
Learning Bounds for Domain Adaptation (2007, methodology): “Several authors have empirically studied a sp ecial case of this in which each instance is weighted separately in the loss function, and instance we ights are set to approximate the target domain distribution [10, 5, 9, 11].”
Sample Selection Bias Correction Theory (2008, methodology): “The analysis of the discriminative method of (Bickel et al., 2007) for the problem of covariate shift could perhaps also benefit from this study.”
Multi-task learning for HIV therapy screening (2008, background): “Bickel et al. (2007) derive a discriminative ptrain(x) expression for this marginal density ratio that can be estimated without estimating the potentially highdimensional densities of training and test distributions by discriminating training against test data.”
Covariate Shift in Hilbert Space: A Solution via Sorrogate Kernels (2013, background): “For example, in (Bickel et al., 2007), a discriminative model (MAP classifier) is proposed that directly characterizes the divergence between the training and test distributions.”
Domain Generalization and Adaptation Using Low Rank Exemplar SVMs (2018, methodology): “density ratios) based on different strategies and criteria in the literature, which includes Kernel Mean Machting (KMM) [21], the MaxNet based methods [22], KUllbackLeibler Importance Estimation Procedure (KLIEP) [23], the Logistic Regression based method (LogReg) [24], and Large Scale KLIEP (LS-KLI”
Feature Weighting for Improved Classifier Robustness (2009, methodology): “The state of the art methods such as those proposed in [11, 2] infer the correcting weight without the need of explicit density estimation.”