Aditya Chattopadhyay

Ph.D. Candidate, Johns Hopkins University. Email: achatto1 at jhu dot edu dot com


Office: 322, Clark Hall, JHU

Baltimore, MD, 21218

About me

I am currently a Ph.D. candidate in the Computer Science Department at the Johns Hopkins University (JHU). My Ph.D. advisors are Prof. René Vidal and Prof. Donald Geman.
Ph.D. dissertation topic: An Information-theoretic Framework for Designing Interpretable Predictors.

Before JHU, I did my M.S. in Computational Natural Sciences and B.Tech in Computer Science from the International Institute of Information Technology, Hyderabd (IIIT-H). My M.S. advisor was Prof. U. Deva Priyakumar.
M.S. dissertation topic: A probabilistic framework for constructing temporal relations in Monte Carlo trajectories.

Research Interests

My current research involves developing Machine Learning (ML) algorithms that not only make accurate predictions but also provide human-interpretable explanations for their predictions (the users of these algorithms may not be ML experts themselves). Given that ML algorithms are increasingly being deployed in critical decision-making areas such as finance and healthcare, ensuring the interpretability of these algorithms is essential. For more information about my work, please visit the Projects section.

Apart from developing ML algorithms, I am also keenly interested in their application in biology. In the past, I worked on developing ML techniques to understand the unfolding of biomolecules (proteins + RNAs) from their molecular dynamic trajectories. To know more about this work, visit the Projects section.


Jan 15, 2024 Our paper on an interpretable-by-design framework for image classification using Large Language and Vision models integrated with Information Pursuit (a greedy algorithm that makes predictions by sequentially asking informative questions about the input image) accepted as a conference paper at ICLR 2024!
Sep 22, 2023 Our paper on an information theoretic perspective of Orthogonal Matching Pursuit and its applications to explainable AI was accepted at NeurIPS 2023 as a spotlight presentation! (acceptance rate: 3%).
Sep 16, 2023 Presenting a poster at DeepMath 2023.
Jan 21, 2023 Our paper on an information theoretic framework for making interpretable predictions (explainable-AI) was accepted as a journal paper at TPAMI, with a follow-up more efficient version accepted as a conference paper at ICLR 2023!
Sep 18, 2022 Presenting a poster at DeepMath 2022.
May 6, 2021 Honoured to receive the MINDS Fellowship for Spring and Summer semesters in 2021.
Sep 9, 2019 Our work on explainable-AI advocating using causality to interpret neural networks via post-hoc attribution has been featured by Hindustan Times. Check out the article here :sparkles: :smile:.

Selected Publications

  1. ICLR
    Bootstrapping Variational Information Pursuit with Large Language and Vision Models for Interpretable Image Classification
    Aditya Chattopadhyay, Kwan Ho Ryan Chan, and Rene Vidal
    In The Twelfth International Conference on Learning Representations, 2024
  2. NeuRIPS
    Information Maximization Perspective of Orthogonal Matching Pursuit with Applications to Explainable AI
    Aditya Chattopadhyay, Ryan Pilgrim, and Rene Vidal
    In Thirty-seventh Conference on Neural Information Processing Systems, 2023
  3. TPAMI
    Interpretable by design: Learning predictors by composing interpretable queries
    Aditya Chattopadhyay, Stewart Slocum, Benjamin D Haeffele, and 2 more authors
    IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022
  4. ICLR
    Variational Information Pursuit for Interpretable Predictions
    Aditya Chattopadhyay, Kwan Ho Ryan Chan, Benjamin David Haeffele, and 2 more authors
    In The Eleventh International Conference on Learning Representations, 2023