Yujia Bao 包昱嘉

Building 32 G484, MIT

yujia@csail.mit.edu

I am a fourth-year PhD student at MIT CSAIL, advised by Prof. Regina Barzilay. Before I came to MIT, I spent five years studying math in Shanghai and Madison. I got my bachelor degree from Shanghai Jiao Tong University in 2016, and my master degrees from UW-Madison in 2017 and from MIT in 2018.

I became interested in machine learning back in 2016, when I was working with Prof. David Page and Prof. Rebecca Willett on Hawkes process modeling of adverse drug reactions [MLHC17]. My recent reserach focuses on the following topics:

  • Transfer learning: improving low-resource performance by connecting human rationale and model attention [EMNLP18]; leveraging distributional signatures for few-shot text classification [ICLR20].
  • Robustness (against spurious correlations): learning stable classifiers by comparing different data environments [ICML21] and by transferring spurious features [Preprint].

news

Jun 15, 2021 Our work Learning Stable Classifiers by Transferring Unstable Features is now available on arXiv.
May 8, 2021 Our work Predict then Interpolate: A Simple Algorithm to Learn Stable Classifiers was accepted to ICML 2021.
Dec 19, 2019 Our work Few-shot Text Classification with Distributional Signatures was accepted to ICLR 2020.

selected publications

  1. Learning Stable Classifiers by Transferring Unstable Features
    Yujia Bao, Shiyu Chang, and Regina Barzilay
    Preprint 2021
  2. Predict then Interpolate: A Simple Algorithm to Learn Stable Classifiers
    Yujia Bao, Shiyu Chang, and Regina Barzilay
    In International Conference on Machine Learning 2021
  3. Few-shot Text Classification with Distributional Signatures
    Yujia Bao*, Menghua Wu*, Shiyu Chang, and Regina Barzilay
    In International Conference on Learning Representations 2020
  4. Deriving Machine Attention from Human Rationales
    Yujia Bao, Shiyu Chang, Mo Yu, and Regina Barzilay
    In Empirical Methods in Natural Language Processing 2018