Timezone: »

Detecting and Correcting for Label Shift with Black Box Predictors
Zachary Lipton · Yu-Xiang Wang · Alexander Smola

Fri Jul 13 02:50 AM -- 03:00 AM (PDT) @ A3

Faced with distribution shift between training and test set, we wish to detect and quantify the shift, and to correct our classifiers without test set labels. Motivated by medical diagnosis, where diseases (targets), cause symptoms (observations), we focus on label shift, where the label marginal p(y) changes but the conditional p(x| y) does not. We propose Black Box Shift Estimation (BBSE) to estimate the test distribution p(y). BBSE exploits arbitrary black box predictors to reduce dimensionality prior to shift correction. While better predictors give tighter estimates, BBSE works even when predictors are biased, inaccurate, or uncalibrated, so long as their confusion matrices are invertible. We prove BBSE's consistency, bound its error, and introduce a statistical test that uses BBSE to detect shift. We also leverage BBSE to correct classifiers. Experiments demonstrate accurate estimates and improved prediction, even on high-dimensional datasets of natural images.

Author Information

Zachary Lipton (Carnegie Mellon University)
Yu-Xiang Wang (UC Santa Barbara)
Yu-Xiang Wang

Yu-Xiang Wang is the Eugene Aas Assistant Professor of Computer Science at UCSB. He runs the Statistical Machine Learning lab and co-founded the UCSB Center for Responsible Machine Learning. He is also visiting Amazon Web Services. Yu-Xiang’s research interests include statistical theory and methodology, differential privacy, reinforcement learning, online learning and deep learning.

Alexander Smola (Amazon)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors