Sohil Shah
"If I have the belief that I can do it, I shall surely acquire the capacity to do it, even if I may not have it at the beginning" - Mahatma Gandhi
As a Research Manager at Avataar.ai, a Series B startup headquartered in India, I direct a team pioneering AI algorithms that transform visual experiences across e-commerce platforms.
Earlier, I was a Research Scientist at Intel's Embodied AI Lab (previously Intelligent Systems Lab under Vladlen Koltun). Prior to this, I received my Doctorate from department of ECE at University of Maryland, College Park in 2018. My thesis titled "Optimization Algorithms Using Priors in Computer Vision" was co-advised by Prof. Tom Goldstein and Prof. Larry S. Davis. The thesis won ECE's Distinguished Dissertation Award - given to best dissertation for the year 2018. Since 2017, I continue to serve as reviewer for - TPAMI, CVPR, ECCV, ICCV, NeurIPS, ICLR, ICML, & AAAI.
During the course of my PhD at UMD, I was fortunate to intern at Xerox Labs and Intel Labs. Earlier, I worked as a Systems Design, Staff-I Engineer at Broadcom Corporation in Mobile and Wireless group and a Design Engineer at Beceem Communications Pvt. Ltd in PHY-SW group. I obtained my Masters Degree in Electrical Engineering from IIT Bombay, India. During my masters, I was advised by Prof. Vikram Gadre at TI-DSP Research Lab. I graduated with bachelor's degree in Electrical and Communication Engineering from MIT, Anna University.
My research interest lies at the intersection of theoretical Machine Learning and practical Computer Vision. The problems in 2D & 3D Computer Vision and Computationally Biology involving Generative Models, Graph Modeling, Numerical Optimization and Deep Neural Network interest me.
UPDATES
Received Outstanding Reviewer Award for CVPR 2019!
Joined Intel Labs as Research Scientist.
Attending ICLR 2018 at Vancouver, BC.
Defended PhD Dissertation successfully. My dissertation presentation slides can be downloaded from here.
Selected as winner of University of Maryland ECE Distinguished Dissertation Fellowship Award (2018). (given to outstanding PhD students in the ECE department)
One paper accepted at ICLR 2018.
The code for Robust Continuous Clustering is now online.
My work on Clustering with Vladlen Koltun from Intel Labs is accepted for publication in PNAS 2017.
Starting April 2017 - Advanced to candidacy.
Xerox awarded Travel grant of $2000 for ECCV 2016.
US Patent Application 15225936 filed for "Method and System for Multimedia Processing to identify Concepts in Multimedia."
Two papers accepted at ECCV 2016.
One paper accepted at CVPR 2016.
Summer and Fall 2016 - Interned at Intel Visual Computing Labs at Santa Clara.
Summer 2015 - Interned at Xerox Research Center India, Bangalore.
RESEARCH / PUBLICATIONS
Molecular Conformation
Sohil Shah, and Vladlen Koltun, "Conformation Generation using Transformer Flows", 2022.
Graph Synthesis
Sohil Shah, and Vladlen Koltun, "Auto-decoding Graphs", 2020.
Semantic Segmentation
Sohil Shah, Pallabi Ghosh, Larry Davis and Tom Goldstein, "Stacked U-Nets: A No-frills Approach to Natural Image Segmentation", 2018. Code
Arthita Ghosh, Max Ehrlich, Sohil Shah, Larry Davis and Rama Chellappa, "Stacked U-Nets for Ground Material Segmentation in Remote Sensing Imagery", In CVPR DeepGlobe Workshop, 2018.
Generative Adversarial Networks (GANs)
Sohil Shah*, Abhay Yadav*, Zheng Xu, David Jacobs, and Tom Goldstein, "Stabilizing Adversarial Nets with Prediction Methods", In ICLR 2018. Code,Poster
Ryen Krusinga, Sohil Shah, Matthias Zwicker, Tom Goldstein and David Jacobs, "Understanding the (un)interpretability of natural image distributions using generative models", 2019.
Continuous Clustering
Sohil Shah and Vladlen Koltun, "Robust Continuous Clustering", In PNAS (Proceeding of National Academy of Science) Journal, August 2017. Code
Sohil Shah and Vladlen Koltun, "Deep Continuous Clustering", 2018. Code
Video Understanding
Sohil Shah, Kuldeep Kulkarni, Arijit Biswas, Ankit Gandhi, Om Deshmukh and Larry Davis, "Weakly Supervised Learning of Heterogeneous Concepts in Videos", In ECCV 2016. Code, Poster
Ankit Gandhi, Arijit Biswas, Om Deshmukh, Sohil Shah, Kuldeep Kulkarni, "Method and system for multimedia processing to identify concepts in multimedia", US Patent App. 15/225,936
Semidefinite Programming
Sohil Shah*, Abhay Yadav*, David Jacobs, Carlos Castillo, Christoph Studer and Tom Goldstein,"Biconvex Relaxation for Semidefinite Programming in Computer Vision", In ECCV 2016. Code, Poster
Block Sparsity
Sohil Shah, Tom Goldstein and Christoph Studer, "Estimating Sparse Signal with Smooth Support via Convex Programming and Block Sparsity", In CVPR 2016. Code,Poster
Other publications
Shah S.A., Venkatesan P., Sundar D., Kannan M., “Low latency, High throughput and Less complex VLSI architecture for 2D-DFT”, In Proceeding of IEEE International Conference on Signal Processing, Communication and Networking, Jan 4-6, 2008, pp. 349-353.
Sohil Atul Shah and M. Ganesh Madhan, “Distortion Analysis of VCSEL for Radio over Fiber (ROF) Transmission”, In Proceeding of 33rd OSI Symposium on Optics and Optoelectronics, Dec 2007.
Technical Reports
Master Thesis, Applications of Filterbanks and Wavelets in Error Control Codes, June 2010.
Undergrad Research, Design and Analysis of OFDM based Radio over Multimode Fibre system for Wireless LAN, May 2008.
Throughput Optimal Control of Energy Harvesting Networks, July 2013.