Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment
By Yifan Wu, Ezra Winston, Divyansh Kaushik, Zachary Chase Lipton. 2019
133 citations recorded by the source corpus.
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. they can break down under shifting label distributions. We propose asymmetrically-relaxed distribution alignment, a new approach that overcomes some limitations of standard domain-adversarial algorithms. Moreover, we characterize precise assumptions under which our algorithm is theoretically principled and demonstrate empirical benefits on both synthetic and real datasets.
Subfields: Representation & Self-Supervised Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Representation & Self-Supervised Learning
Citations in the corpus
- Learning Robust Global Representations by Penalizing Local Predictive Power (2019, background): “While some methods have justified domain-adversarial learning by appealing to theoretical bounds due to Ben-David et al. (2010a), the theory does not in fact guarantee generalization (recently shown by Johansson et al. (2019) and Wu et al. (2019)) and sometimes guarantees failure.”
- Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems (2020, methodology): “BEAR (Kumar et al., 2019a) uses the MMD distance to represent the policy constraint, while Wu et al. (2019a) evaluates an instantiation of the first order Wasserstein distance.”
- Behavior Regularized Offline Reinforcement Learning (2019, background): “…well to MMD even though they are not designed for support matching; (ii) we briefly tried divergences that are explicitly designed for support matching (the relaxed KL and relaxed Wasserstein distances proposed by Wu et al. (2019)), but did not observe a clear benefit to the additional complexity.”
- Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction (2019, methodology): “Perhaps methods from outside RL, predominantly used in domain adaptation, such as using asymmetric f-divergences [38] can be used for support restriction.”
- WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “…et al., 2014; Taghipour and Ng, 2016), estimating student knowledge and ability (Desmarais and Baker, 2012; Wu et al., 2020), identifying students who need assistance (Ahadi et al., 2015), or automatically generating explanations for student submissions (Williams et al., 2016; Wu et al., 2019).”