I’m a Ph.D. student at MIT in the Clinical and Applied Machine Learning Group. I work on building machine learning models that are resilient to unreliable data. This interest map summarizes questions I care about.
I went to MIT for undergrad and majored in computer science with a concentration in South Asian studies. I’m also a former intern at Microsoft Research, Borealis AI, D.E. Shaw Research, Counsyl and Aetion. You can find my CV here or email me at divyas at mit dot edu, or both!
I’m interested in how we can deploy and develop machine learning models in the presence of unreliable data. Within this area, I’m interested in machine learning on noisy data, model uncertainty, and dataset shift. Right now, I’m curious about the value of data augmentation during inference.
Unsupervised Domain Adaptation in the Absence of Source Data
R. Sahoo*, D.Shanmugam*, J. Guttag
Uncertainty & Robustness in Deep Learning Workshop, ICML 2020
Automated Image Segmentation of Liver Stage Malaria Infection
AP. Soleimany, H. Suresh, JJG. Ortiz, D. Shanmugam, N. Gural, J. Guttag, SN. Bhatia
Computational Biology Workshop ICML 2019
Disparities in the performance of Natural Language Processing Tools
H. Suresh, D. Shanmugam
Women in Machine Learning Workshop 2017
When and Why Test-Time Augmentation Works
D. Shanmugam, D. Blalock, G. Balakrishnan, J. Guttag
Learning to Limit Data Collection for Data Minimization
D. Shanmugam, S. Shabanian, F. Diaz, M. Finck, A. Biega
Addressing Feature-Dependent Label Noise: A Generative Framework
J. Sahota, D. Shanmugam, J. Ramanan, S. Eghbali, M. Brubaker
MIT News, “Using machine learning to estimate the risk of cardiovascular death”, Rachel Gordon, September 12, 2019