University of California, Santa Cruz
I’m an Assistant Professor of Computer Science and Engineering at UC Santa Cruz. My research interests are crowdsourcing and algorithmic fairness, both in the context of machine learning. The central question associated with my work is learning from dynamic and noisy data.
Previously I was holding a postdoctoral fellow position at Harvard University. I have a Ph.D. from the University of Michigan, Ann Arbor and a B.Sc. from Shanghai Jiao Tong University, China.
My research is generously supported by the National Science Foundation, Office of Naval Research, Amazon (in collaboration with NSF FAI), and UC Santa Cruz. I was partially supported by the DARPA SCORE program.
noisylabels.com is online! We collected and published re-annotated versions of the CIFAR-10 and CIFAR-100 data which contains real-world human annotation errors. We show how these noise patterns deviate from the classically assumed ones and what the new challenges are. We hope these datasets will facilitate the benchmarking and development of weakly supervised learning solutions.
[Join us!] I am looking for highly-motivated postdocs, phd students, visitors, and interns to work with us on weakly-supervised learning and algorithmic fairness. If you are interested, please email me at .
[2022.01] We have two papers accepted to ICLR 2022. It is official now noisylabels.com will be presented at ICLR this year. Our second acceptance reveals the disparate impact of popular semi-supervised learning algorithms on different sub-populations of data. And yes, “rich does get richer”: details here.
[2022.01] Our paper Fair Classification with Instance-dependent Label Noise is accepted to the inaugural CLeaR 2022! We provide a causally inspired approach for training a fair classifier when training data suffers from complex label noise.
[2021.12] I’m serving FAccT 2022 and UAI 2022 as the Area Chair.
[2021.11] Our paper Unfairness Despite Awareness: Group-Fair Classification with Strategic Agents is accepted and selected as a spotlight at StratML@NeurIPS 2021. We observe a fairness reversal phenomenon, where a trained-to-be-fair classifier might in fact lead to more unfair treatments when agents are strategically responding.
[2021.11] I’m invited to give a talk at StratML@NeurIPS 2021.
[2021.10] Our group has 4 papers accepted to NeurIPS 2021 with one spotlight selection! These selected works span the study of fairness in machine learning and learning from weak supervisions. We have shown possible persistent qualification disparity from careless deployment of models (spotlight!), delayed impacts of actions in bandit learning, as well as improving fairness guarantee when learning from noisy labels. We have also defined the problem of weakly supervised policy learning.
[2021.09] Our work Are Gender-Neutral Queries Really Gender-Neutral? Mitigating Gender Bias in Image Search has been accepted to EMNLP 2021 for an oral presentation.
[2021.08] Our paper Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels has won the best paper award at IJCAI 2021 workshop on Weakly Supervised Representation Learning. Congrats to Zhaowei and Yiwen (our former summer intern)!
[2021.08] I gave an invited talk at IJCAI 2021 workshop on Weakly Supervised Representation Learning.
[2021.07] Our paper Linear Classifiers that Encourage Constructive Adaptation has won the best paper award at ICML 2021 workshop on Algorithmic Recourse. Congrats to Yatong and Jialu!
[2021.07] Congrats to Yatong who has received the BSOE inaugural Fellowship for Anti-Racism Research for her research on algorithmic fairness!
[2021.06] I’ll be serving as the Area Chair for NeurIPS 2021.
[2021.05] My new preprint (ICML 2021 long talk forthcoming) highlights the disparate effect of memorizing instance-dependent noisy labels. I also show how several existing learning with noisy label solutions fare at instance level.
[2021.05] We provide a new tool to estimate noise transition matrix (to appear at ICML 2021). Check it out! It is efficient, model-free, scalable, and can be broadly applied in a learning with noisy label task. Technical details here.
[2021.01] Our proposal Fairness in Machine Learning with Human in the Loop is now awarded by the NSF FAI program! Thank you NSF & Amazon! As the lead institute, we will receive $1M to conduct a wide range of research on understanding the interaction of machine learning and human agents.