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 activities are funded by the Centre for Basic Machine Learning Research in Life Sciences.

Most recently, I was interning at Microsoft Research in London and working with SOTA generative models for robust image/video manipulation.

Before that, I was in New York collaborating with Rajesh Ranganath and Mark Goldstein at NYU. We are collaborating on effective sampling techniques, controllability and modelling discrete data. Pre-print coming soon!

My 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.

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 controllable generation and alternative guidance methods, 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

• [2025/10] I am back in Copenhagen!
• [2025/8] I am looking for research roles in various sectors. Get in touch!
• [2025/6] I’ve made it to London! Starting at Microsoft soon.
• [2025/5] NeurIPS submission sprint!