About me

Hi, I am a fourth-year PhD student at the Technical University of Denmark at the Applied Mathematics and Computer Science Department. I am supervised by Jes Frellsen at DTU and Wouter Boomsma at DIKU(University of Copenhagen). My research is a part of the Centre for Basic Machine Learning Research in Life Sciences.

My recent internship at SonyAI with the Music Foundational Model team was fruitful! Interested in non-linear diffusion models, controllability and disentanglement? Read about our work here.

Currently, I am in New York collaborating with Rajesh Ranganath and Mark Goldstein at NYU. I am working on effective sampling techniques, controllability and modelling discrete data.

I am looking for internship roles staring in April-May 2025 for four to six months. Please reach out if interested.

Click for a longer bio. Prior to starting this graduate program, I was a research associate at RBCDSAI, IIT Madras and before that I finished my integrated masters in ECE with a specialisation in signal processing and pattern recognition from IIIT Bangalore. During my masters, in 2019 I also spent some time in the Approximate Bayesian Inference team at RIKEN-AIP in Tokyo working under Emtiyaz Khan on scaling Bayesian natural gradient optimizers.

Research Interests

Lately, I have been focusing on interesting problems in generative modelling. Specifically, I like to think about current challenges with Diffusion Models, like non-linear inference processes, simplifying guidance in pre-trained models, accelerating sampling and modelling discrete data. I also enjoy dabbling in structured modelling and scientific applications of diffusion models.

I have broad interests in the field of probabilistic modelling. I am enthusiastic about problems in Bayesian deep learning related to uncertainty quantification, variational inference and high-dimensional sampling. More generally, I like pondering about generalisation, distribution shifts and identifiability in deep probabilistic models. I wish I was better at differential geometry, measure theory and high-dimensional sampling. Works that aim to quantify or estimate uncertainty, model densities, or infer underlying structures/phenomena probabilistically, intrigue me.

Feel free to reach out! (Email: ansup at dtu dot dk)

News

  • [2025/2] Pre-print of my CVPR work is out, check the publications tab! Work done at Sony AI.
  • [2024/12] I am looking for research interships!
  • [2024/11] Submitted to CVPR
  • [2024/10] I have made it to NYC!
  • [2024/6] I am heading to Tokyo, will be working with the DGM team at SonyAI for a few months!
Old updates.
  • [2022/11] Personal Website created.
  • [2023/04] ICML 2023 submission rejected!
  • [2023/09] NeurIPS submission accepted! My work on scaling implicit variational approximations has been accepted as a Spotlight!
  • [2023/12] Presented my work at NeurIPS 2023 in New Orleans and enjoyed some exquisite live jazz.
  • [2024/3] I have been offered research internships!
  • [2024/5] I was at ICLR in Vienna, beautiful city with a lot of history, amazing conference with many good orals, the VAE paper got the test of time award!
  • [2024/5] Failed to submit to NeurIPS.