Timezone: »

Pareto Domain Adaptation
fangrui lv · Jian Liang · Kaixiong Gong · Shuang Li · Chi Harold Liu · Han Li · Di Liu · Guoren Wang

Wed Dec 08 12:30 AM -- 02:00 AM (PST) @

Domain adaptation (DA) attempts to transfer the knowledge from a labeled source domain to an unlabeled target domain that follows different distribution from the source. To achieve this, DA methods include a source classification objective to extract the source knowledge and a domain alignment objective to diminish the domain shift, ensuring knowledge transfer. Typically, former DA methods adopt some weight hyper-parameters to linearly combine the training objectives to form an overall objective. However, the gradient directions of these objectives may conflict with each other due to domain shift. Under such circumstances, the linear optimization scheme might decrease the overall objective value at the expense of damaging one of the training objectives, leading to restricted solutions. In this paper, we rethink the optimization scheme for DA from a gradient-based perspective. We propose a Pareto Domain Adaptation (ParetoDA) approach to control the overall optimization direction, aiming to cooperatively optimize all training objectives. Specifically, to reach a desirable solution on the target domain, we design a surrogate loss mimicking target classification. To improve target-prediction accuracy to support the mimicking, we propose a target-prediction refining mechanism which exploits domain labels via Bayes’ theorem. On the other hand, since prior knowledge of weighting schemes for objectives is often unavailable to guide optimization to approach the optimal solution on the target domain, we propose a dynamic preference mechanism to dynamically guide our cooperative optimization by the gradient of the surrogate loss on a held-out unlabeled target dataset. Our theoretical analyses show that the held-out data can guide but will not be over-fitted by the optimization. Extensive experiments on image classification and semantic segmentation benchmarks demonstrate the effectiveness of ParetoDA

Author Information

fangrui lv (Beijing Institute of Technology)
Jian Liang (Alibaba Group)

Jian Liang received his Ph.D. degree from Tsinghua University, Beijing, China, in 2018. During 2018 and 2020 he was a senior researcher in the Wireless Security Products Department of the Cloud and Smart Industries Group at Tencent, Beijing. In 2020 he joined the AI for international Department, New Retail Intelligence Engine, Alibaba Group as a senior algorithm engineer. His paper received the Best Short Paper Award in 2016 IEEE International Conference on Healthcare Informatics (ICHI).

Kaixiong Gong (Beijing Institute of Technology, Tsinghua University)
Shuang Li (Beijing Institute of Technology)
Chi Harold Liu (Beijing Institute of Technology)
Han Li (Alibaba Group)
Di Liu
Guoren Wang

More from the Same Authors