Semi-Supervised Learning of Class Balance under Class-Prior Change by Distribution Matching

By Marthinus Christoffel du Plessis, Masashi Sugiyama. 2012

139 citations recorded by the source corpus.

In real-world classification problems, the class balance in the training dataset does not necessarily reflect that of the test dataset, which can cause significant estimation bias. If the class ratio of the test dataset is known, instance re-weighting or resampling allows systematical bias correction. However, learning the class ratio of the test dataset is challenging when no labeled data is available from the test domain. In this paper, we propose to estimate the class ratio in the test dataset by matching probability distributions of training and test input data. We demonstrate the utility of the proposed approach through experiments.

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. Bernhard Schölkopf
2. Kun Zhang
3. Wen Li
4. Mingming Gong
5. Lixin Duan
6. Dominik Janzing
7. Yahao Liu
8. Jinhong Deng

In Signal Processing

1. Bernhard Schölkopf
2. Kun Zhang
3. Mingming Gong
4. Jessa Bekker
5. Jesse Davis
6. Lixin Duan
7. Wen Li
8. Dominik Janzing

Citations in the corpus

• Multi-Source Domain Adaptation: A Causal View (2015, background): “(Plessis and Sugiyama 2012) and (Zhang et al. 2013) studied how to estimate the change in P Y in this situation, and the latter also applies for regression problems (i.e., with continuous Y ).”
• Undoing the Damage of Label Shift for Cross-domain Semantic Segmentation (2022, background): “Some approaches [6, 19] try to address this problem, but they typically assume that the conditional distribution is the same across domains, which is not realistic in practice.”
• Learning from positive and unlabeled data: a survey (2018, methodology): “However, recently a matching method has been proposed to estimate the class prior when this is not the case [81].”
• A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts (2023, unknown): “as minimizing the KL divergence between the target data distribution pT (x) and its approximation by a linear combination of class-wise distributions pT (x) [554], [555], pT (x) = ∑”