Yujia Bao



๐Ÿง‘๐Ÿปโ€๐Ÿ”ฌ Lead machine learning scientist @ Insitro
๐Ÿพ A proud owner of samoyeds
๐Ÿ› Completed Ph.D. at MIT CSAIL
๐Ÿƒโ€โ™‚๏ธ Staying active in Cambridge, MA


We need Efficient AI

๐Ÿš€ connecting human rationales with model attention for low-resource learning1
๐Ÿš€ meta learning over distributional signatures for few-shot classification2


We need Robust AI

๐Ÿ›ก identifying spurious correlations by contrasting different data environments3
๐Ÿ›ก transferring the knowledge of biases across tasks4
๐Ÿ›ก learning to split any dataset so that predictors cannot generalize across the splits5


Want to know more? Check out my thesis defense on YouTube.




Learning is a lifelong journey.


๐ŸŽ“ B.S. in Mathematics & Applied Mathematics, Shanghai Jiao Tong University, 2016.
๐ŸŽ“ M.A. in Mathematics, University of Wisconsin-Madison, 2017.
๐ŸŽ“ S.M. in Computer Science, MIT, 2019.
๐ŸŽ“ Ph.D. in Computer Science, MIT, 2022 (advised by Regina Barzilay).
๐Ÿง Continue thinking.




References


  1. Yujia Bao, Shiyu Chang, Mo Yu, Regina Barzilay. โ€œDeriving Machine Attention from Human Rationales.โ€ EMNLP 2018.
  2. Yujia bao, Menghua Wu, Shiyu Chang, Regina Barzilay. โ€œFew-shot Text Classification with Distributional Signatures.โ€ ICLR 2020.
  3. Yujia Bao, Shiyu Chang, Regina Barzilay. โ€œPredict then Interpolate: A Simple Algorithm to Learn Stable Classifiers.โ€ ICML 2021.
  4. Yujia Bao, Shiyu Chang, Regina Barzilay. โ€œLearning Stable Classifiers by Transferring Unstable Features.โ€ ICML 2022.
  5. Yujia Bao, Regina Barzilay. โ€œLearning to Split for Automatic Bias Detection.โ€ arXiv 2022.