Maximum Mean Discrepancy for Generalization in the Presence of Distribution and Missingness Shift
This addresses generalization issues in predictive modeling for real-world applications with distribution and missingness shifts, representing an incremental improvement.
The paper tackles covariate shift in predictive modeling by minimizing Maximum Mean Discrepancy (MMD) between training and test sets, proposing three techniques (MMD Representation, MMD Mask, MMD Hybrid) for different shift scenarios, resulting in improved performance, calibration, and extrapolation on the test set.
Covariate shifts are a common problem in predictive modeling on real-world problems. This paper proposes addressing the covariate shift problem by minimizing Maximum Mean Discrepancy (MMD) statistics between the training and test sets in either feature input space, feature representation space, or both. We designed three techniques that we call MMD Representation, MMD Mask, and MMD Hybrid to deal with the scenarios where only a distribution shift exists, only a missingness shift exists, or both types of shift exist, respectively. We find that integrating an MMD loss component helps models use the best features for generalization and avoid dangerous extrapolation as much as possible for each test sample. Models treated with this MMD approach show better performance, calibration, and extrapolation on the test set.