Ayan Sinha

I am a postdoctoral reseacher at MIT, where I work on computational imaging. I did my PhD at Purdue where I developed machine learning techniques inspired by physics. I did my masters at Georgia Tech where I worked on managing uncertainty in complex systems. Prior to that I did my undergrad at I.I.T. Kharagpur.

Email  /  CV  /  Research Statement  /  Google Scholar  /  LinkedIn

Its me!
Research

My research aims to answer the most pressing problems of "Big Data" by combining physical laws and machine learning. I have developed fast and reliable algorithms to uncover the underlying patterns in the data. The `secret sauce' in these techniques is inspired by physical laws and processes like heat flow, Gauss's law and biological neural networks applied to graphs. To test the robustness, I have implemented these techniques to a wide variety of domains by representing data points and their pairwise relationships as a graph structure without rebuilding them for each domain.

feedback

Deconvolving feedback loops in recommender systems
Ayan Sinha, David Gleich, Karthik Ramani
Neural Information Processing Systems (NIPS), 2016
supplement / bibtex / code

We propose a way to remove feedback effects in recommender systems and uncover the true preferences of items by users.


feedback

Deep learning 3D shape surfaces using geometry images
Ayan Sinha, Jing Bai, Karthik Ramani
European Conference of Computer Vision (ECCV), 2016
supplement / bibtex / code

We develop and learn a geometry image representation of a 3D shape surface to ease deep learning of 3D shapes.


feedback

DeepHand: Robust Hand Pose Estimation by Completing a Matrix Imputed with Deep Features
Ayan Sinha, Chiho Choi, Karthik Ramani
Computer Vision and Pattern Recognition (CVPR), 2016
supplement / bibtex / video

We track and reconstruct the hand in realtime using a combination of deep learning and matrix completion techniques.


feedback

A Collaborative Filtering Approach to Real-Time Hand Pose Estimation
Chiho Choi, Ayan Sinha, Joon Hee Choi, , Sujin Jang, Karthik Ramani
International Conference of Computer Vision (ICCV), 2015
supplement / bibtex

We develop a realtime hand pose estimation system inspired from a recommender system.


feedback

Multiscale kernels using random walks
Ayan Sinha, Karthik Ramani
Symposium of Geometry Processing (SGP), 2013
bibtex

We develop multiscale kernels for 3D shape surfaces and unify shape signature representation.


feedback

Uncertainty management in design of complex systems
Ayan Sinha, Karthik Ramani
Journal of Mechanical Design (JMD), 2013
bibtex

We develop a methodology to manage uncertainty by mediating information economics and robust design techniques.


Under Review
feedback

Gauss's law for networks directly reveals community boundaries
Ayan Sinha, David Gleich, Karthik Ramani
Nature Communications
bibtex

We develop a network analog of Gauss's law for electrostatics to identify communites in networks.


feedback

Learning to generate 3D shape surfaces using deep residual networks
Ayan Sinha, Asim Unmesh, Qixing Huang, Karthik Ramani, George Barbastathis
Computer Vision and Pattern Recognition (CVPR)
bibtex

We train deep residual networks to generate both manmade and organic 3D shape surfaces.


Teaching
pacman

Introduction to Mechanical Engineering Design-ME263