Srinivas Venkattaramanujam

I am a MSc student at Mila, where I work on Reinforcement Learning advised by Prof. Doina Precup

In the past I worked on Automatic Speech Recognition with Prof. S. Umesh at IIT Madras and before that I worked as a software engineer at PickYourTrail and Verizon.

Email  /  CV  /  LinkedIn  /  Twitter

Research

I'm interested in Reinforcement Learning and more broadly sequence prediction. Much of my recent work has been on devising algorithms for eliminating the need for domain knowledge for RL tasks.

Self-supervised Learning of Distance Functions for Goal-Conditioned Reinforcement Learning
Srinivas Venkattaramanujam, Eric Crawford, Thang Doan, Doina Precup
Under Review in ICML, 2020
Teaching

COMP 767: Reinforcement Learning - Winter'19 (TA)
Instructor: Prof. Doina Precup

COMP 202: Foundations of Programming - Fall'18 (TA)
Instructor: Dr. Giulia Alberini and Dr. Joseph P Vybihal

COMP 303: Software Design - Fall'17 (TA)
Instructor: Prof. Martin Robillard

Course Projects

COMP 598 (Fall'18): A Guided Tour of 'Metrics for MDPs with Infinite State Spaces'
Provided the background and explained the ideas of Metrics for Markov Decision Processes with Infinite State Spaces

COMP 767 (Winter'18): A reimplementation of TreeQN
Provided a reimplementation of TreeQN in TreeQN and ATreeC: Differentiable Tree-Structured Models for Deep Reinforcement Learning and later fixed a (minor) bug in the authors' public implementation.

IFT 6269 (Fall'17): Implementation of Latent Dirichlet Allocation (Team of 3)
Provided a reimplementation of LDA using Gibbs Sampling and Variational Inference

Work Projects

Speech Lab(IITM): Long Audio Alignment using Kaldi
Create smaller audio segments along with its corresponding transcription from the longer audio segment and its transcription. This toolkit is functionally similar to SailAlign and implements ideas from these two papers.

PickYourTrail: Travel Genie (Android app)
Developed the Android app which displayed the digital itineraries of the users and provided notifications before each activity. The layout and icons were designed by a highly talented designer, Ajit, and Hari. I also contributed to the first version of Veho.

ShopperLane: Android app, Inventory management and billing web apps
Built ShopperLane with the vision of democratizing the inventory information of neighbourhood stores. Products are searchable across stores within the specified radius of a user's location or alternatively can be browsed within a store.

Useful Links

You and Your Research

Life Stories