I'm a first-year PhD in Computer Science student at Stanford University currently rotating in the Kundaje Lab. Previously, I did research in the Azizi Lab at Columbia University and the Pe'er Lab at Memorial Sloan Kettering Cancer Center. My research lies at the intersection of machine learning and cancer genomics. Specifically, I'm interested in developing novel machine learning methods which can both learn from data and incorporate prior knowledge, and applying these methods to drive biological discovery and enable precision medicine.
I also love teaching! Most recently, I taught for AI4ALL, a CS/AI summer program for high school students at Columbia. My goal is to teach students mad skills and get them excited about CS, so that we can increase diversity in our field.
Cameron Park, Shouvik Mani, Satyen Gohil, Katie Maurer, Catherine J. Wu, Elham Azizi
International Conference on Machine Learning (ICML) 2022, Workshop on Computational Biology
[journal submission under preparation]
Cell-cell interactions are fundamental to normal biological processes
and disease, but their evolution over time is poorly understood. DIISCO characterizes
the temporal dynamics of intercellular interactions using scRNA-seq data
from multiple time-points. It features a Bayesian framework which infers interactions
between cell types according to their co-evolution and incorporates
prior knowledge on receptor-ligand complexes.
Shouvik Mani, Michael A. Haddad, Dan Constantini, Willy Douhard, Qiwei Li, Louis Poirier
Computer Vision and Pattern Recognition (CVPR) 2020, Diagram Image Retrieval and Analysis Workshop [supplemental][video][blog post]
A computer vision pipeline which digitizes engineering diagrams by detecting
common symbols (red), detecting and parsing text (blue), and identifying
connections between symbols via lines (not shown).
Shouvik Mani, Mehdi Maasoumy, Sina Pakazad, Henrik Ohlsson
Neural Information Processing Systems (NeurIPS) 2019, Learning with Rich Experience Workshop
Distance Metric Learning Regularization (DMLreg) is an approach to elicit
prior knowledge from domain experts through pairwise similarity comparisons and
incorporate that knowledge into a regularized linear model. DMLreg helps improve
model performance in high-dimensional settings.
One of my favorite memories from CMU is participating in
Buggy is a CMU tradition where student orgs build carbon fiber vehicles and train
year-round to prepare for relay races on a mile-long course during Spring Carnival.
And yes, there is a driver inside the vehicle!
Pushing Buggy for a CMU student org
I also like traveling (and flying) a lot! I hope to get my pilot's license one day. Some cool places I've been to: