Dwaipayan Saha

Hey! I primarily work on foundational models for alternative data modalities (tabular, matrix, time-series, etc), efficient policy optimization via reinforcement learning, and algorithms that leverage underlying data structure. Prior to this, I did some work at the intersection of high-dimensional statistics and econometrics.

I am a PhD student at Columbia University advised by Anish Agarwal.

Before that, I graduated Summa Cum Laude from Princeton University with a B.S.E. in Computer Science and Math. Here, I was fortunate to be advised by Matt Weinberg. I received the department’s Outstanding Student Teaching Award for my TA contributions across several undergraduate and graduate courses.

Two of my favorite courses at Princeton were Stochastic Calculus and Advanced Algorithm Design. A sizeable chunk of my time here was spent thinking about theoretical computer science and probability theory all of which I still love chatting about!

Feel free to reach out at ds4386 (at) columbia (dot) edu. Here is my CV.

news

Aug 28, 2026	Wrapped up my summer at Quadrature!
Jan 23, 2026	Finished my internship at Hudson River Trading!
Dec 06, 2025	Presented TabImpute: Universal Zero-Shot Imputation for Tabular Data at AI for Tabular Data Workshop @ EurIPS 2025 in Copenhagen, Denmark!
Oct 03, 2025	Poster presentation on recent work TabImpute: Universal Zero-Shot Imputation for Tabular Data at CAIRFI 2025.
Sep 29, 2025	Started a deep learning alpha research internship at Headlands Technologies!

latest posts

Aug 10, 2024	Variational Autoencoders and the EM Algorithm

selected publications

Foundational Models

MatrixFormer: A Foundation Model for Matrix Completion

Dwaipayan Saha^*, Jacob Feitelberg^*, Kyuseong Choi^* , and 2 more authors

In Submission, 2026

arXiv Bib

@article{saha2026matrixformerfoundationmodelmatrix,
  title = {MatrixFormer: A Foundation Model for Matrix Completion},
  author = {Saha, Dwaipayan and Feitelberg, Jacob and Choi, Kyuseong and Dwivedi, Raaz and Agarwal, Anish},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  journal = {In Submission},
}

Kernels & ML

Manifold Random Features

Dwaipayan Saha^*, Ananya Parashar^*, Derek Long^* , and 1 more author

NeurIPS, 2026

arXiv Bib

@article{parashar2026manifoldrandomfeatures,
  title = {Manifold Random Features},
  author = {Saha, Dwaipayan and Parashar, Ananya and Long, Derek and Choromanski, Krzysztof},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  journal = {NeurIPS},
}

Optimal Transport

Near-Linear Time Generalized Sinkhorn Algorithms for Bounded Genus Graphs

Krzysztof Choromanski, Derek Long, Ananya Parashar , and 1 more author

NeurIPS, 2026

arXiv Bib

@article{choromanski2026nearlineartimegeneralizedsinkhorn,
  title = {Near-Linear Time Generalized Sinkhorn Algorithms for Bounded Genus Graphs},
  author = {Choromanski, Krzysztof and Long, Derek and Parashar, Ananya and Saha, Dwaipayan},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.DS},
  journal = {NeurIPS},
}

RL & Post-Training

GOPO: Policy Optimization using Ranked Rewards

Dwaipayan Saha^*, Kyuseong Choi^*, Woojeong Kim^* , and 2 more authors

COLM Sci-Fi, 2026

arXiv Bib

@article{choi2026gopopolicyoptimizationusing,
  title = {GOPO: Policy Optimization using Ranked Rewards},
  author = {Saha, Dwaipayan and Choi, Kyuseong and Kim, Woojeong and Agarwal, Anish and Dwivedi, Raaz},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  journal = {COLM Sci-Fi},
}

Foundation Models

TabImpute: Universal Zero-Shot Imputation for Tabular Data

Jacob Feitelberg, Dwaipayan Saha, Kyuseong Choi , and 3 more authors

AI for Tabular Data Workshop @ EurIPS, 2025

arXiv Bib

@article{feitelberg2025tabimputeaccuratefastzeroshot,
  title = {TabImpute: Universal Zero-Shot Imputation for Tabular Data},
  author = {Feitelberg, Jacob and Saha, Dwaipayan and Choi, Kyuseong and Ahmad, Zaid and Agarwal, Anish and Dwivedi, Raaz},
  year = {2025},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  journal = {AI for Tabular Data Workshop @ EurIPS},
}

Stat & CI

Synthetic Blips: Generalizing Synthetic Controls for Dynamic Treatment Effects+

Anish Agarwal, Sukjin Han, Dwaipayan Saha , and 2 more authors

In Submission, 2025

arXiv Bib

@article{agarwal2025syntheticblipsgeneralizingsynthetic,
  title = {Synthetic Blips: Generalizing Synthetic Controls for Dynamic Treatment Effects+},
  author = {Agarwal, Anish and Han, Sukjin and Saha, Dwaipayan and Syrgkanis, Vasilis and Yoon, Haeyeon},
  year = {2025},
  archiveprefix = {arXiv},
  primaryclass = {econ.EM},
  journal = {In Submission},
}