Selected Research

This section covers a few selected papers from my years as an AI researcher in healthcare and an information theorist. The synopses expressed here are my own and do intend to speak on behalf of my collaborators or employers.

You can find my full publication list on Google Scholar.

AI for Medical Imaging

• ELIXR: Towards a general purpose x-ray artificial intelligence system through alignment of large language models and radiology vision encoders. ArXiv, abs/2308.01317, 2023.

  Synopsis: This work introduces ELIXR, an approach to combining a language-aligned radiology image model with a fixed large language model (LLMs) to perform a broad range of tasks. In addition to state-of-the art performance on data-efficient classification tasks and semantic search, ELIXR shows promise for new capabilities, including visual question-answering and radiology report quality assurance.
  Team: S. Xu, L. Yang, C.J. Kelly, M. Sieniek, T. Kohlberger, M.Q. Ma, W-H Weng, A.P. Király, S. Kazemzadeh, Z. Melamed, J. Park, P. Strachan, Y. Liu, C. Lau, P. Singh, C. Chen, M. Etemadi, S.R. Kalidindi, Y. Matias, K. Chou, G.S. Corrado, S. Shetty, D. Tse, S. Prabhakara, D. Golden, R. Pilgrim, K. Eswaran, and A. Sellergren.

• Chest Radiograph Interpretation with Deep Learning Models: Assessment with Radiologist-adjudicated Reference Standards and Population-adjusted Evaluation. Radiology, page 191—293, 12 2019. ISSN 0033-8419. doi: 10.1148/radiol. 2019191293. URL http://pubs.rsna.org/doi/10.1148/radiol.2019191293.

  Synopsis: This work demonstrates the potential of deep learning models to detect clinically relevant findings in chest X-rays by producing a high quality, radiologist-adjudicated reference standard. In addition to training and evaluating a model against this reference standard, we also reference standard labels for validation and test sets on the NIH Chest X-ray dataset.
  Team: A. Majkowska, S. Mittal, D.F. Steiner, J. Reicher, S.M. McKinney, G.E. Duggan, K. Eswaran, P-H.C. Chen, Y. Liu, S.R. Kalidindi, A. Ding, G.S. Corrado, D. Tse, and S. Shetty.

AI for Environmental Health, Social Determinants of Health

• General Geospatial Inference with a Population Dynamics Foundation Model. ArXiv, abs/2411.07207, 2024.
  Synopsis: This work trains a graph neural network model that captures rich dependencies between a given location and human behavior from maps, busyness, aggregated search trends, and environmental factors such as weather and air quality. The resulting model shows state-of-the-art performance on a suite of 27 downstream tasks covering three distinct domains: health indicators, socioeconomic factors, and environmental measurements.
  Team: M. Agarwal, M. Sun, C. Kamath, A. Muslim, P. Sarker, J. Paul, H. Yee, M. Sieniek, K. Jablonski, Y. Mayer, D. Fork, S. de Guia, J. McPike, A. Boulanger, T. Shekel, D. Schottlander, Y. Xiao, M.C. Manukonda, Y. Liu, N. Bulut, A. Abu-el-haija, B. Perozzi, M. Bharel, V. Nguyen, L. Barrington, N. Efron, Y. Matias, G. Corrado, K. Eswaran, S. Prabhakara, S. Shetty, G. Prasad
  Tags: Environmental Health, Social Determinants of Health, Graph Neural Networks

• Multimodal LLMs for health grounded in individual-specific data. In Workshop on Machine Learning for Multimodal Healthcare Data, pages 86–102. Springer, 2023.
  Synopsis: This work explores how large language models can be used as general-purpose multimodal predictors, combining structured demographic data, high dimensional measures of lung function, and lab results, to predict disease risk over a range of phenotypes. Our approach, HeLM, matches the performance of the best models for predicting cataracts, GERD, Exczema, Osteoarthritis, and Pneumonia, without being explicitly trained on the trait. The model is evaluated on the UKBiobank dataset, a large-scale biomedical database and research resource containing de-identified genetic, lifestyle and health information from the United Kingdom.
  Team: A. Belyaeva, J. Cosentino, F. Hormozdiari, K. Eswaran, S. Shetty, G. Corrado, A. Carroll, C.Y. McLean, and N.A. Furlotte.
  Tags: Personalized Health, UKBiobank, Large Language Models

Information Theory, Reinforcement Learning, Security, and Statistical Signal Processing

• Remote Source Coding under Gaussian Noise: Dueling Roles of Power and Entropy Power. IEEE Trans. Inf. Theory, Oct. 2018.
  Synopsis: This work extends and generalizes a result that shows that denoising algorithms relying on distributed quantization of noisy observation data experience a performance loss that is exponentially worse in the number of distributed sensors when compared to a system that can centrally coordinate to produce sufficient statistics of the noisy observation data prior to quantization. While this result was previously shown when the sources and noise have jointly Gaussian distributions, our work extends this result even when the underlying sources are non-Gaussian.
  Team: K. Eswaran and M. Gastpar
  Tags: Statistical Signal Processing, Information Theory

• Feedback communication and control over a single channel. IEEE Trans. Inf. Theory, vol. 59, no. 10, pp. 6243-6257, Oct. 2013.
  Synopsis: This paper develops a theoretical framework to understand the fundamental limits of error control coding under dynamic constraints, e.g. a low-cost wireless communication system that can be pre-empted by a higher tier service using the same spectrum. The formulation ties concepts in information theory to those of stochastic control and reinforcement learning, uncovering a tradeoff between mutual information and the Bellman equation.
  Team: K. Eswaran and M. Gastpar
  Tags: Reinforcement Learning, Information Theory

• Secrecy via sources and channels. IEEE Transactions on Information Theory, vol. 58, no. 11, pp. 6747 – 6765, Nov. 2012.
  Synopsis: This paper develops a theoretical framework to explore information-theoretically secure communication protocols that are provably resistant to the attacks, including those made possible by quantum computers. The approach shows how a communication protocol can be designed to exploit known statistical properties of the communication channel as well as shared context between the sender’s source and receiver side information to generate information-theoretically secure secret keys.
  Team: V. M. Prabhakaran, K. Eswaran, and K. Ramchandran
  Tags: Security, Information Theory