Xi Peng

Leading XLearning Group at the College of Computer Science, Sichuan University. From 2014 to 2017, he was a research scientist at A*STAR Singapore. He received his Ph.D. degree from Sichuan University in 2013.

In these areas, he has tried to provide novel insights to communities, and some major results are briefly highlighted below:

• Proving the equivalence between F-norm (L2) and nuclear-norm (L1) under mild conditions. The theoretical results well explain why L2-norm-based representations are competitive with L1-norm-based ones in number of applications (TNNLS'16, TCYB'17a, TCYB'17b).
• PARTY: one of the FIRST deep clustering methods (arxiv'15, IJCAI'16), which is proposed to group the complex data in an unsupervised way.
• TELL: the FIRST interpretable unsupervised neural network which is intrinsically explainable and transparent (JMLR'22).
• CC: it achieves state of the art for online clustering in an elegant contrastive learning fashion (AAAI'21, IJCV'22).
• Noisy Correspondence (NC): A new research DIRECTION of noisy labels learning. NC aims to address the mismatched pairs instead of annotation errors in a number of tasks, e.g., fundamental models, retrieval, Re-ID, QA, dialogue, graph matching, identification, etc. (CVPR'21, NeurIPS'21, CVPR'22, CVPR'23, TPAMI'22, ACMMM'22, AAAI'23, etc).
• All in One Image/Video Restoration and Enhancement: for the first time, we explored this new direction which aims to handle multiple degradations with one mode (CVPR'21).
• AI4LifeScience: My lab also attempts to explore the core problems in interdiscipline and develop specific solutions. Currently, we are diving into AI4LifeScience (Nature Communications 2023).