Debottam Dutta

I am a Ph.D. student in Electrical and Computer Engineering at the University of Illinois Urbana-Champaign, advised by Prof. Romit Roy Choudhury in the Signals and Inference Research Group (SiNRG). My research centers on generative models(Diffusion, Flows and EBMs) with an emphasis on conditional and efficient generation. Currently I am exploring compositional abilities (the ability to represent and combine multiple semantic concepts) of these models, and how corrective or guided sampling can improve their alignment and reliability.

Before my Ph.D., I worked as a Research Fellow at the LEAP Lab at Indian Institute of Science, advised by Prof. Sriram Ganapathy, where I studied representation learning and interpretability for audio and speech.

News

Jan 29, 2026	CO3 got accepted at ICLR 2026
Jan 18, 2025	Our paper on Curvature Guided Monte-Carlo got accepted at ICASSP 2025
Sep 30, 2024	A paper got accepted at NeurIPS 2024 Workshop on AI-Driven Speech, Music, and Sound Generation.
Jan 1, 2024	Journal paper got accepted for TASLP.
Jun 22, 2023	Coswara dataset paper got accept at Nature Scientific data!

Selected Publications

ICLR

Steer Away From Mode Collisions: Improving Composition In Diffusion Models

Debottam Dutta, Jianchong Chen, Rajalaxmi Rajagopalan, Yu-Lin Wei, and Romit Roy Choudhury

2026

ICLR 2026

Bib PDF Code Website

@misc{dutta2026steerawaymodecollisions,
  title = {Steer Away From Mode Collisions: Improving Composition In Diffusion Models},
  author = {Dutta, Debottam and Chen, Jianchong and Rajagopalan, Rajalaxmi and Wei, Yu-Lin and Choudhury, Romit Roy},
  year = {2026},
  note = {ICLR 2026},
  eprint = {2509.25940},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
}

ICASSP

Estimating Multi-chirp Parameters using Curvature-guided Langevin Monte Carlo

Sattwik Basu, Debottam Dutta, Yu-Lin Wei, and Romit Roy Choudhury

In ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2025

Bib PDF

@inproceedings{10888413,
  author = {Basu, Sattwik and Dutta, Debottam and Wei, Yu-Lin and Roy Choudhury, Romit},
  booktitle = {ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  title = {Estimating Multi-chirp Parameters using Curvature-guided Langevin Monte Carlo},
  year = {2025},
  volume = {},
  number = {},
  pages = {1-5},
  keywords = {Monte Carlo methods;Chirp;Signal processing algorithms;Linear programming;Polynomials;Reliability;Noise measurement;Speech processing;Optimization;Signal to noise ratio;Chirps;Optimization;Parameter Estimation;Langevin Monte Carlo (LMC);Gaussian smoothing;Curvature},
  doi = {10.1109/ICASSP49660.2025.10888413},
}

Neurips

Multi-Source Music Generation with Latent Diffusion

Zhongweiyang Xu, Debottam Dutta, Yu-Lin Wei, and Romit Roy Choudhury

In Audio Imagination: NeurIPS 2024 Workshop AI-Driven Speech, Music, and Sound Generation 2024

Bib PDF

@inproceedings{xu2024multisource,
  title = {Multi-Source Music Generation with Latent Diffusion},
  author = {Xu, Zhongweiyang and Dutta, Debottam and Wei, Yu-Lin and Choudhury, Romit Roy},
  year = {2024},
  booktitle = {Audio Imagination: NeurIPS 2024 Workshop AI-Driven Speech, Music, and Sound Generation},
  url = {https://openreview.net/forum?id=bxzUnfWrgk},
}