Documents
Poster
Poster
Robust importance-weighted cross-validation under sample selection bias
- Citation Author(s):
- Submitted by:
- Wouter Kouw
- Last updated:
- 11 October 2019 - 4:48pm
- Document Type:
- Poster
- Document Year:
- 2019
- Event:
- Presenters:
- Wouter Kouw
- Paper Code:
- 27
- Categories:
- Log in to post comments
Cross-validation under sample selection bias can, in principle, be done by importance-weighting the empirical risk. However, the importance-weighted risk estimator produces sub-optimal hyperparameter estimates in problem settings where large weights arise with high probability. We study its sampling variance as a function of the training data distribution and introduce a control variate to increase its robustness to problematically large weights.