Recently, the top international journal, IEEE Transactions on Pattern Analysis and Machine Intelligence（TPAMI）, accepted a research paper entitled "Towards Accurate and Robust Domain Adaptation Under Multiple Noisy Environments" from professor Yin Yilong's team. IEEE TPAMI is the top journal in artificial intelligence and machine learning, recommended by the Chinese Computer Society (CCF) as a Class A journal. It is a JCR I Top journal of the Chinese Academy of Sciences. The latest impact factor (IF) is 24.31. Prof. Yin Yilong is the sole corresponding author of the paper, his PhD student Han Zhongyi is the first author, and Shandong University is the first author and corresponding author institute. This is the first time for graduate students of the School of Software to publish academic papers in TPAMI as the first author.

In many non-stationary environments, machine learning algorithms usually confront distribution shift scenarios. Previous domain adaptation methods have achieved great success. However, they would lose algorithm robustness in multiple noisy environments where the examples of source domain become corrupted by label noise, feature noise, or open-set noise. This paper reports an attempt toward achieving noise-robust domain adaptation. This paper first gives a theoretical analysis and finds that different noises have disparate impacts on the expected target risk. To eliminate the effect of source noises, this paper proposes offline curriculum learning minimizing a newly-defined empirical source risk. This paper suggests a proxy distribution-based margin discrepancy to gradually decrease the noisy distribution distance to reduce the impact of source noises. This paper proposes an energy estimator for assessing the outlier degree of open-set-noise examples to defeat the harmful influence. This paper also suggests robust parameter learning to mitigate the negative effect further and learn domain-invariant feature representations. Finally, this paper seamlessly transforms these components into an adversarial network that performs efficient joint optimization for them. A series of empirical studies on the benchmark datasets and the COVID-19 screening task show that the proposed algorithm remarkably outperforms the state-of-the-art, with over 10% accuracy improvements in some transfer tasks.

In recent years, Prof. Yin Yilong's team has focused on robust domain adaptation theories and methods in open environments, focusing on fundamental research and original innovation, and working intensively around the bottleneck of scientific problems and national needs. The related research results have been published in CCF A journals IEEE TPAMI, TIP, and CCF A conferences IJCAI 2020, AAAI 2022, CVPR 2022, ACMMM 2022.

Link to the article: https://ieeexplore.ieee.org/abstract/document/9921307