I am currently a research scientist at Google Brain. I finished my PhD at MIT in 2017, where I was advised by Prof. Regina Barzilay and a member of NLP group at MIT CSAIL. I obtained my bachelor of science degree from Peking University in 2010.
Prior to joining Google, I was a scientist and research leader at ASAPP, where I manage a group of researchers and engineers for applied machine learning and natural language processing.
Broadly speaking, I am interested in the algorithmic perspective of machine learning and various applications of natural language processing. My recent research focuses on the following topics:
-
Efficient ML: devise neural network models that can scale at training or inference while retaining state-of-the-art results on target applications (EMNLP21, EMNLP18, ICML17); and build novel learning algorithms for model compression (EMNLP20a, EMNLP20b).
-
Rationalizing model prediction: introduce self-explaining models that selects a subset of input as a justification for the models prediction (EMNLP16, ACL20a); or thru interacting with the user (ACL20b).
News
| 04/11/23 | We released our paper on CoDA, a new adaptation method for large models in NLP, vision and speech. CoDA achieves 2-8x inference speedup with modest to no loss of accuracy. |
|---|---|
| 03/17/23 | We released our paper on CoLT5, an encoder-decoder architecture for long inputs using conditional computation, achieving new SOTA on the SCROLLS benchmark. |
| 10/29/21 | SRU++ receives Outstanding Paper award at EMNLP 2021. |
| 02/24/21 | We released SRU++, a highly efficient RNN that obtains top results but also reduce training cost significantly. [News] [Paper] |
| 09/14/20 | Our team got 3 long papers accepted at EMNLP 2020. |