Vivekkumar Patel

I am a graduate student in the department of Computer Science at Stanford University. I am primarily interested in Deep Learning and its application to Computer Vision, Natural Language Processing and Reinforcement Learning.

I am currently working on Battery life Prediction models within the Ermon group under the supervision of Prof. Stefano Ermon.

I did my undergraduation from Indian Institute of Technology Bombay with a major in Electrical Engineering and minor in Computer Science.

Email  /  LinkedIn  /  Resume

Internships
project_img

Google Brain, Mountain View

Worked on Recommender Systems with the SIR team in Google Brain.

Developed a novel set of algorithms for Top-k optimization in recommender systems.

project_img

Daikin, Shiga, Japan

Worked in the Devices group for software development of residential air-conditioners.

Developed a system and implemented the FxLMS algorithm to implement Active Noise Cancellation.

Notable Projects
project_img

Neural Techniques for Pose Guided Image Generation

We implemented the PG^2 model to generate image of a given person in a given pose.

Modified the algorithm to stabilize and speed up the training.

Poster Report Code

project_img

Machine Comprehension on SQuAD

We re-implemented the BiDAF model from scratch.

Analysed the benefits of different components and suggested ways to improve the performance.

Poster Report

project_img

Robust Deep RL for Autonomous driving

Implemented Deep Deterministic Policy Gradient (DDPG) algorithm to autonomously drive a race car on the TORCS simulator.

Checked it's robustness to additive white Gaussian noise, and suggested ways to make it more robust.

Poster Report

project_img

Deep Reinforcement Learning for Atari Games

Implemented DQN from scratch for two Atari games: Space Invaders and Q*bert.

Also implemented the Double and Dueling DQN architectures and compared the performance.

Poster Report

project_img

Recommender Systems

Built a graph based recommender system for non-binary rating predictions.

Graph based algorithms scale linearly with the size of data and perform better than matrix factorization when working with limited computational resources.

Poster Report


inspired from this website