Research / Peer reviewed
The balancing principle for parameter choice in distance-regularized domain adaptation
Werner Zellinger, Natalia Shepeleva, Marius-Constantin Dinu, Hamid Eghbal-zadeh, Duc Hoan Nguyen, Bernhard Nessler, Sergei V. Pereverzyev, Bernhard A. Moser
- Venue
- NeurIPS
- Year
- 2021
- Citations
- 19
- Length
- 14 pages
Abstract
We address the unsolved algorithm design problem of choosing a justified regularization parameter in unsupervised domain adaptation. This problem is intriguing as no labels are available in the target domain. Our approach starts with the observation that the widely-used method of minimizing the source error, penalized by a distance measure between source and target feature representations, shares characteristics with regularized ill-posed inverse problems. Regularization parameters in inverse problems are optimally chosen by the fundamental principle of balancing approximation and sampling errors. We use this principle to balance learning errors and domain distance in a target error bound. As a result, we obtain a theoretically justified rule for the choice of the regularization parameter. In contrast to the state of the art, our approach allows source and target distributions with disjoint supports. An empirical comparative study on benchmark datasets underpins the performance of our approach.
Cite this paper
@inproceedings{zellinger2021balancing,
author = {Zellinger, Werner and Shepeleva, Natalia and Dinu, Marius-Constantin and Eghbal-zadeh, Hamid and Nguyen, Duc Hoan and Nessler, Bernhard and Pereverzyev, Sergei V. and Moser, Bernhard A.},
title = {The balancing principle for parameter choice in distance-regularized domain adaptation},
booktitle = {NeurIPS},
year = {2021},
url = {https://www.dinu.at/research/the-balancing-principle-for-parameter-choice-in-distance-regularized-dom},
}Werner Zellinger, Natalia Shepeleva, Marius-Constantin Dinu, Hamid Eghbal-zadeh, Duc Hoan Nguyen, Bernhard Nessler, Sergei V. Pereverzyev, Bernhard A. Moser. (2021). The balancing principle for parameter choice in distance-regularized domain adaptation. NeurIPS. https://www.dinu.at/research/the-balancing-principle-for-parameter-choice-in-distance-regularized-dom
TY - CPAPER AU - Zellinger, Werner AU - Shepeleva, Natalia AU - Dinu, Marius-Constantin AU - Eghbal-zadeh, Hamid AU - Nguyen, Duc Hoan AU - Nessler, Bernhard AU - Pereverzyev, Sergei V. AU - Moser, Bernhard A. TI - The balancing principle for parameter choice in distance-regularized domain adaptation T2 - NeurIPS PY - 2021 UR - https://www.dinu.at/research/the-balancing-principle-for-parameter-choice-in-distance-regularized-dom ER -
More peer reviewed
- CoLLAs · 2024 SymbolicAI: A framework for logic-based approaches combining generative models and solvers
- Ph.D. Thesis, Johannes Kepler University · 2024 Parameter Choice and Neuro-Symbolic Approaches for Deep Domain-Invariant Learning
- ICLR · 2023 Addressing Parameter Choice Issues in Unsupervised Domain Adaptation by Aggregation
- CoLLAs · 2022 A Dataset Perspective on Offline Reinforcement Learning