Timezone: »
In this paper, we study the problem of domain adaptation regression, which learns a regressor for a target domain by leveraging the knowledge from a relevant source domain. We start by proposing a distribution-informed neural network, which aims to build distribution-aware relationship of inputs and outputs from different domains. This allows us to develop a simple domain adaptation regression framework, which subsumes popular domain adaptation approaches based on domain invariant representation learning, reweighting, and adaptive Gaussian process. The resulting findings not only explain the connections of existing domain adaptation approaches, but also motivate the efficient training of domain adaptation approaches with overparameterized neural networks. We also analyze the convergence and generalization error bound of our framework based on the distribution-informed neural network. Specifically, our generalization bound focuses explicitly on the maximum mean discrepancy in the RKHS induced by the neural tangent kernel of distribution-informed neural network. This is in sharp contrast to the existing work which relies on domain discrepancy in the latent feature space heuristically formed by one or several hidden neural layers. The efficacy of our framework is also empirically verified on a variety of domain adaptation regression benchmarks.
Author Information
Jun Wu (University of Illinois, Urbana Champaign)
Jingrui He (University of Illinois at Urbana-Champaign)
Sheng Wang
Kaiyu Guan (University of Illinois at Urbana-Champaign)
Dr. Kaiyu Guan is a Blue Waters Associate Professor in ecohydrology and remote sensing at the University of Illinois at Urbana-Champaign (UIUC). Guan got his B. Sci. from Nanjing University, M.A. and PhD from Princeton University with Eric Wood, and conducted postdoc research in Stanford University with David Lobell and Joe Berry. Guan’s group at UIUC focuses on bringing the interdisciplinary domain knowledge (plant ecology, hydrology, biogeochemistry, and climate science), satellite/airborne data, fieldwork, supercomputing, and machine learning together to revolutionize how we monitor and model plant-water-nutrient interactions for agricultural ecosystems, across the U.S. and globe. His group’s work aims to increase our society’s resilience and adaptability to maintain sustainability of ecosystem services, food security and water resources under the influence of climate change and anthropogenic drivers. Guan serves as PI and Co-PI for 15+ federal grants from NASA, NSF, DOE, and USDA. Guan has published 90+ peer-reviewed papers in leading scientific journals. Guan is the awardee of NSF CAREER Award, NASA New Investigator Award, AGU Early Career Award in Global Environmental Change, Hyperion Research High-Performance-Computing Innovation Excellence Award, SoAR Foundation’s National Selection of U.S. Agricultural Research, FFAR Seeding Solution Award, etc.
Elizabeth Ainsworth (University of Illinois at Urbana-Champaign)
More from the Same Authors
-
2022 Poster: Augmentations in Hypergraph Contrastive Learning: Fabricated and Generative »
Tianxin Wei · Yuning You · Tianlong Chen · Yang Shen · Jingrui He · Zhangyang Wang -
2022 Poster: Deep Active Learning by Leveraging Training Dynamics »
Haonan Wang · Wei Huang · Ziwei Wu · Hanghang Tong · Andrew J Margenot · Jingrui He -
2022 Poster: Improved Algorithms for Neural Active Learning »
Yikun Ban · Yuheng Zhang · Hanghang Tong · Arindam Banerjee · Jingrui He -
2012 Poster: GenDeR: A Generic Diversified Ranking Algorithm »
Jingrui He · Hanghang Tong · Qiaozhu Mei · Boleslaw K Szymanski -
2007 Spotlight: Nearest-Neighbor-Based Active Learning for Rare Category Detection »
Jingrui He · Jaime Carbonell -
2007 Poster: Nearest-Neighbor-Based Active Learning for Rare Category Detection »
Jingrui He · Jaime Carbonell