About me: I am a Ph.D. candidate in Statistics at the University of Michigan working with Prof. Yuekai Sun and Prof. Moulinath Banerjee. Prior to joining Michigan, I completed my undergraduate and masters studies at Indian Statistical Institute, Kolkata with a focus on mathematics and statistics.Â
Research interests: I am broadly interested in the mathematical foundations of data science. In the course of my PhD researches I have devoted a majority of my time in the areas of transfer learning and algorithmic fairness. Some of my recent interests are:
Transfer learning models,
Fairness and Interpretability in ML algorithms,
Integrative analysis on heterogeneous data.
E-mail: smaity (at) umich (dot) edu
Recent works:
Maity, S., Yurochkin, M., & Sun, Y. (2023+). An investigation of allocation and representation harms in contrastive learning. (under review in ICLR 2024) [preprint]
Maia Polo, F.*, Maity, S.*, Banerjee M., Sun, Y., & Yurochkin, M. (2023+). Estimating Fréchet bounds for validating programmatic weak supervision. (under review in ICLR 2024).
Ngweta, L.*, Maity, S.*, Agarwal, M., Gittens, A., Sun, Y., & Yurochkin, M. (2023+). Aligners: Decoupling LLMs and Alignment. (draft in progress).
Maity, S., Roy, S., Xue, S., Yurochkin, M., & Sun, Y. (2023+). How does overparametrization affect performance on minority groups?. (to be submitted). [preprint]
Maity, S., Dutta, D., Terhorst, J., Sun, Y., & Banerjee, M. (2023). A linear adjustment based approach to posterior drift in transfer learning. Biometrika, 2023. [paper]
Maity, S., Yurochkin, M., Banerjee, M., & Sun, Y. (2023). Understanding new tasks through the lens of training data via exponential tilting. In The Eleventh International Conference on Learning Representations, 2023. [paper]
Maity, S.*, Mukherjee, D.*, Banerjee, M., & Sun, Y. (2023). Predictor-corrector algorithms for stochastic optimization under gradual distribution shift. In The Eleventh International Conference on Learning Representations, 2023. [paper]
Bakshi, S.*, & Maity, S.* (2023). Bayes classifier cannot be learned from noisy responses with unknown noise rates. Accepted as Tiny paper in the Eleventh International Conference on Learning Representations, 2023. [paper]
Ngweta, L., Maity, S., Gittens, A., Sun, Y., & Yurochkin, M. (2023). Simple disentanglement of style and content in visual representations. In Fortieth International Conference on Machine Learning, 2023. [paper]
Professional Experiences:
Graduate Research Assistant (Winter 2020 - present): Dept. of Statistics, University of Michigan, Ann Arbor, MI.
Research Intern (Summer 2022): IBM Research, Cambridge, MA.
Research Assistant (Summer 2017): Johns Hopkins School of Public Health, Baltimore, MD.
Research Intern (Summer 2016): Johns Hopkins School of Public Health, Baltimore, MD.
Teaching Experiences:
STATS 250, Undergrad level intro. to Statistics, Fall 2018.
STATS 470, Design of experiments, Fall 2019.
STATS 430, Intro. to stochastic processes, Winter 2020.Â