Vansh Bansal

prof_pic.jpg

vansh at utexas.edu

I am a third-year doctoral student in the Department of Statistics and Data Sciences at UT Austin, where I am fortunate to be advised by James Scott and Purnamrita Sarkar, and collaborate with Alessandro Rinaldo. My research focuses at the intersection of high-dimensional statistics and deep learning, with an interest in developing computationally efficient sampling algorithms and applying them to key problems in deep generative models, including validation, guidance, and uncertainty quantification.

Previously, I completed my bachelor’s degree in the Department of Computer Science and Engineering at the Indian Institute of Technology, Kanpur, where I had the privelege to be advised by Dootika Vats. I also had the opportunity to work with the amazing Piyush Rai, Vipul Arora and Ashutosh Modi.

Beyond work, I am an Indian classical vocalist and I enjoy going for hikes.

PS: I’d try my best to give my advice and feedback to those who are applying to graduate programs in statistics, machine learning or related fields, particularly to those for whom this type of feedback would usually be unavailable. The best way to reach out to me is by email.

news

Mar 11, 2026 I have advanced to candidacy for my Ph.D. at UT Austin! My presentation focused on bridging statistical theory with practical algorithms to validate, guide, and accelerate deep generative models.
Mar 01, 2026 Our paper titled “Score-Guided Proximal Projection: A Unified Geometric Framework for Rectified Flow Editing” got accepted at ICLR 2026 (DeLTA workshop).
Jan 23, 2026 Our paper titled “On the convergence and straightness of Rectified Flows” got accepted at AISTATS 2026.
Oct 17, 2025 Selected as a top reviewer at NeurIPS 2025.
Sep 18, 2025 We got a spotlight presentation (top 3%) at NeurIPS 2025 for the paper “CoLT: The conditional localization test for assessing the accuracy of neural posterior estimates”! See you in San Diego!

selected publications

  1. ICLR 2026 (DeLTA workshop)
    Score-Guided Proximal Projection: A Unified Geometric Framework for Rectified Flow Editing
    Vansh Bansal, and James G. Scott
    2026
  2. NeurIPS 2025
    CoLT: The conditional localization test for assessing the accuracy of neural posterior estimates
    Tianyu Chen*Vansh Bansal*, and James G. Scott
    2025
  3. Preprint
    Straightness of Rectified Flow: A Theoretical Insight into Wasserstein Convergence
    Vansh Bansal*, Saptarshi Roy*, Purnamrita Sarkar, and 1 more author
    2024
  4. Preprint
    The surprising strength of weak classifiers for validating neural posterior estimates
    Vansh Bansal*, Tianyu Chen*, and James G. Scott
    2025
  5. AISTATS 2025
    Conditional diffusions for neural posterior estimation
    Tianyu Chen, Vansh Bansal, and James G. Scott
    2024