Shamak Dutta

s7dutta at uwaterloo dot ca

I am a PhD student in Electrical and Computer Engineering at the University of Waterloo, where I am advised by Stephen Smith. My research interests are in controls, optimization, and machine learning, with a focus on designing efficient algorithms for sequential decision making problems that arise in robotics.

Previously, I received a bachelor's in Computer Engineering and a master's in Systems Design Engineering from the University of Waterloo. During my master's, I worked with Bryan Tripp and Graham Taylor at the intersection of neuroscience and deep learning. My thesis is on Correlated Noise in Deep Convolutional Neural Networks. I also worked with Hamid Tizhoosh on image retrieval and with Dana Kulic on behaviour cloning for human motion.

During the summer of 2018, I interned at Preferred Networks in Tokyo, where I worked with Shunta Saito and Masaki Saito on conditional video prediction using generative models. I spent the summer of 2017 at Latent Logic in Oxford, where I worked with Joao Messias and Shimon Whiteson on 2D-3D pose reconstruction from single images.

CV  /  Google Scholar  /  GitHub

Research
Convolutional Neural Networks Regularized by Correlated Noise
Shamak Dutta, Bryan Tripp, Graham Taylor
Conference on Computer and Robot Vision, 2018
bibtex

We show that adding spatially and tuning dependent correlated noise to the activities of units in a neural network helps in regularization and image classification under occlusion.

Barcodes for Medical Image Retrieval Using Autoencoded Radon Transform
H.R. Tizhoosh, Christopher Mitcheltree, Shujin Zu, Shamak Dutta
International Conference on Pattern Recognition, 2016
bibtex

We show that barcodes generated by thresholding the learned representations of autoencoders on the Radon transform of images can be used in medical image search and retrieval.


credits