Direct Density Ratio Estimation for Large-scale Covariate Shift Adaptation

By Yuta Tsuboi, Hisashi Kashima, Shohei Hido, Steffen Bickel, Masashi Sugiyama. 2009

173 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

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

1. Wouter M. Kouw
2. Luc Van Gool
3. Wen Li
4. Dong Xu
5. Masashi Sugiyama
6. Dengxin Dai
7. Marco Loog
8. Qiang Yang

In Signal Processing

1. Wouter M. Kouw
2. Marco Loog
3. Takafumi Kanamori
4. Wen Li
5. Dong Xu
6. Teruyuki Suzuki
7. Hui Xiong
8. Zheng Xu

Citations in the corpus

• A Survey on Metric Learning for Feature Vectors and Structured Data (2013, background): “Therefore, adapting a metric learning algorithm to covariate shift boils down to computing the importance weights, which can be done reliably using unlabeled data (Tsuboi et al., 2008).”
• A Review of Domain Adaptation without Target Labels (2019, methodology): “Following this interpretation, the weights are directly estimated using an optimization procedure where the discrepancy between the weighted source and the target distribution is minimized [97].”
• Generalized Energy Based Models (2020, unknown): “…the energies are parametrized by a compact set Ψ with ψ 7→Eψ continuous over Ψ. Estimating the likelihood is then achieved using i.i.d. samples (Xn)1:N ,(Ym)1:M from P and G (Tsuboi et al., 2009; Sugiyama et al., 2012; Liu et al., 2017): L̂P,G(E)=− 1 N N∑ n=1 E(Xn)−log ( 1 M M∑ m=1 exp(−E(Ym)) ) .”
• Density-ratio matching under the Bregman divergence: a unified framework of density-ratio estimation (2012, methodology): “3 Log-linear models Another popular model choice would be the log-linear model (Tsuboi et al. 2009; Kanamori et al. 2010): r(x; θ , θ0) = exp ( ψ(x) θ + θ0 ) , (18)”
• Discriminative Learning Under Covariate Shift (2009, methodology): “Tsuboi et al. (2008) derive an extension to KLIEP for large-scale applications and reveal a close relationship to kernel mean matching.”
• An introduction to domain adaptation and transfer learning (2018, background): “However, just minimizing this objective with respect to w might cause highly varying or unusually scaled values [190].”
• Domain Generalization and Adaptation Using Low Rank Exemplar SVMs (2018, methodology): “Various approaches have been proposed to learn the weights (a.k.a. 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 R”
• A Unified Framework for Metric Transfer Learning (2017, background): “Previous studies about transfer learning focused on either re-weighting the instances in the source domain [10], [11], [12], [13] to match the distribution of the target domain, or finding a new representation for the instances in the Y.”
• Learning Models with Uniform Performance via Distributionally Robust Optimization (2018, background): “Work in domain adaptation seeks models that receive data from one domain and are tested on a specified target; typical approach is to reweight the distribution P0 to make it “closer” to the known target distribution Ptarget [86, 51, 16, 88, 89, 93].”