Timezone: »

 
Oral
Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment
Yifan Wu · Ezra Winston · Divyansh Kaushik · Zachary Lipton

Thu Jun 13 05:10 PM -- 05:15 PM (PDT) @ Room 102
in Learning Theory »

Domain adaptation addresses the common problem when the target distribution generating our test data drifts from the source (training) distribution. While absent assumptions, domain adaptation is impossible, strict conditions, e.g. covariate or label shift, enable principled algorithms. Recently-proposed domain-adversarial approaches consist of aligning source and target encodings, often motivating this approach as minimizing two (of three) terms in a theoretical bound on target error. Unfortunately, this minimization can cause arbitrary increases in the third term, e.g., these methods are unprincipled under label distribution shift. We propose asymmetrically-relaxed distribution alignment, a new approach that overcomes some limitations of standard domain-adversarial algorithms. We characterize precise assumptions under which our algorithm is theoretically principled and demonstrate empirical benefits on both synthetic and real datasets.

[ Slides] 

Author Information

Yifan Wu (Carnegie Mellon University)
Ezra Winston (CMU MLD)
Divyansh Kaushik (Carnegie Mellon University)
Zachary Lipton (Carnegie Mellon University)

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

More from the Same Authors