Semi-Generative Modelling: Covariate-Shift Adaptation with Cause and Effect Features
The 22nd International Conference on Artificial Intelligence and Statistics, 16-18 April 2019
Artificial Intelligence and Statistics (AISTATS)
MetadataShow full item record
Von Kügelgen, B., Loog, M., & Mey, A. (2019). Semi-Generative Modelling: Covariate-Shift Adaptation with Cause and Effect Features. The 22nd International Conference on Artificial Intelligence and Statistics, 16-18 April 2019, 89 1361-1369. https://doi.org/10.17863/CAM.47092
Current methods for covariate-shift adaptation use unlabelled data to compute importance weights or domain-invariant features, while the final model is trained on labelled data only. Here, we consider a particular case of covariate shift which allows us also to learn from unlabelled data, that is, combining adaptation and semi-supervised learning. Using ideas from causality, we argue that this requires learning with both causes, X_C, and effects, X_E, of a target variable, Y, and show how this setting leads to what we call a semi-generative model, P(Y,X_E|X_C,θ). Our approach is robust to domain shifts in the distribution of causal features and leverages unlabelled data by learning a direct map from causes to effects. Experiments on synthetic data demonstrate significant improvements in classification over purely-supervised and importance-weighting baselines.
This record's DOI: https://doi.org/10.17863/CAM.47092
This record's URL: https://www.repository.cam.ac.uk/handle/1810/300021