Postdoctoral Researcher · Institute for Advanced Study, Technical University of Munich
Generative models · Normalizing flows · Theoretical ML · AI for health

I am a postdoc at IAS, TU Munich, working with Prof. Benedikt Wiestler (AI-IDT Lab) and Prof. Anke Meyer-Baese (Florida State University) on generative models for medical imaging. MCML has a short film on the group's work.

I completed my PhD at the Machine Learning Lab, IIIT Hyderabad, advised by Prof. Girish Varma. The thesis, Fast & Efficient Normalizing Flows and Applications of Image Generative Models, develops invertible convolutional layers that are both theoretically characterized and fast on GPU, then uses generative models in vision and scientific applications.

I am interested in collaborating on energy-based models, normalizing flows / flow matching, and generative models for medical imaging.

Previously: Samsung Research (generative super-resolution) · UIUC / UIC College of Medicine (contactless vital signs) · UNSW / University of Sydney (remote sensing) · B.Tech, HBTU Kanpur

• Invertible generative models — characterizing and accelerating invertible convolutions for normalizing flows (CInC, FInC, Inverse-Flow)
• AI for health — 3D brain MRI synthesis, physics-guided tumor growth, remote photoplethysmography in clinical video
• Vision for science — seed quality assessment, geological mapping from satellite imagery, art restoration

• AISTATS 2025 — Inverse-Flow: parallel backpropagation for inverse convolution, $O(\sqrt{n})$ vs. $O(n^3)$
• COSPAR Outstanding Paper Award for Young Scientists, 2024 — Geo-SAE geological mapping (ASR 2024)
• 1st place, NVIDIA competition @ IEEE ICETCI 2023; 1st place, C4MTS challenge @ NCVPRIPG 2023
• ACM India–IARCS travel grant (AISTATS 2025); iHub-Data PhD fellowship; Oxford MLSS 2024; Climate Change AI Summer School 2024

Full list on Google Scholar and my website.

• Parallel Backpropagation for Inverse of a Convolution with Application to Normalizing Flows
  Sandeep Nagar, Girish Varma
  AISTATS 2025 · PMLR · arXiv · project

• CInC Flow: Characterizable Invertible 3×3 Convolution
  Sandeep Nagar, Marius Dufraisse, Girish Varma
  TPM @ UAI 2021 · OpenReview · project · code

• FInC Flow: Fast and Invertible k×k Convolutions for Normalizing Flows
  Aditya Kallappa, Sandeep Nagar, Girish Varma
  VISAPP 2023 (oral) · arXiv · project

• Remote Sensing Framework for Geological Mapping via Stacked Autoencoders and Clustering
  Sandeep Nagar*, Ehsan Farahbakhsh*, Joseph Awange, Rohitash Chandra
  Advances in Space Research 2024 · COSPAR award
  journal · project

• R2I-rPPG: A Robust Region of Interest Selection for Remote Photoplethysmography to Extract Heart Rate
  Sandeep Nagar, Mustafa Alam, Mark Hasegawa-Johnson, David G. Beiser, Narendra Ahuja
  preprint · arXiv

• Automated Seed Quality Testing System using GAN & Active Learning
  Sandeep Nagar, Prateek Pani, Raj Nair, Girish Varma
  PReMI 2021 · dataset · code

• Adaptation of the Super Resolution SOTA for Art Restoration in Camera Capture Images
  Sandeep Nagar, Abhinaba Bala, Sai Amrit Patnaik
  IEEE ICETCI 2023 · arXiv · IEEE · code

_{Full CV, news, and talks: naagar.github.io}