Anirban Samaddar
Portrait of Anirban Samaddar

Anirban Samaddar

Assistant Computational Mathematician, MCS Division,
Argonne National Laboratory

I am a statistician focused on Bayesian machine learning and its use in the natural sciences. My interests include flow matching, diffusion models, information-theoretic deep learning, and neural architecture search. I work on machine learning applications in physics, climate, and materials science.

Lab Mates

Sandeep Madireddy (supervisor), Yixuan Sun, Adarsha Balaji

Education

Ph.D. in Statistics, Michigan State University
Advisors: Dr. Gustavo de los Campos and Dr. Taps Maiti

Contact

asamaddar@anl.gov · anirbansamaddar00@gmail.com

CV LinkedIn Google Scholar GitHub

Research

  1. Spatiotemporal Forecasting of the Edge Localized Modes in Tokamak Plasmas using Neural Networks
    Anirban Samaddar, Qian Gong, Sandeep Madireddy, Christopher Hansen, Semin Joung, David R. Smith, Yixuan Sun, Fatima Ebrahimi, Prasanna Balaprakash, Andrew Oakleigh Nelson
    Machine Learning: Science and Technology, 2025
  2. Efficient Flow Matching using Latent Variables
    Anirban Samaddar, Yixuan Sun, Viktor Nilsson, Sandeep Madireddy
    International Conference on Artificial Intelligence and Statistics (AISTATS), 2026
  3. REMEDI: Corrective Transformations for Improved Neural Entropy Estimation
    Viktor Nilsson*, Anirban Samaddar*, Sandeep Madireddy, Pierre Nyquist (*equal contribution)
    International Conference on Machine Learning (ICML), 2024
  4. Sparsity-Inducing Categorical Prior Improves Robustness of the Information Bottleneck
    Anirban Samaddar, Sandeep Madireddy, Prasanna Balaprakash, Gustavo de los Campos, Tapabrata Maiti, Ian Fischer
    International Conference on Artificial Intelligence and Statistics (AISTATS), 2023
  5. Bayesian Hierarchical Hypothesis Testing in Large-Scale Genome-Wide Association Study
    Anirban Samaddar, Tapabrata Maiti, Gustavo de los Campos
    Genetics, 2024

Full list on Google Scholar.

Projects

Foundation Modeling for Astro

The goal is to develop state-of-the-art foundation models for joint modeling of multi-modal astronomical datasets. The problem involves pretraining and fine-tuning transformer models to learn interpretable embeddings across modalities.

Nuclear Fusion

The goal is to develop foundation models for detecting and mitigating the risk of plasma instability inside Tokamak reactors. The problem involves spatiotemporal generative modeling of the plasma field.

Honors

Awards

Internships

Teaching

STT 200: Statistical Methods — Recitation Instructor, Michigan State University
Fall 2019, Spring 2019, and 2020
Introduction to statistical methods for undergraduates.

Past Students

About Me

Guitar. I enjoy western rock music and play lead guitar in our band under School of Rock. I perform shows around Knoxville, Tennessee.

Travel. I love hiking in the mountains. My two best hiking experiences were at Pictured Rocks National Lakeshore in Up North, Michigan, with my wife, and climbing Sandakphu in West Bengal.