Indian Institute of Technology (IIT) Kharagpur, India |
July 2012 - May 2016
|
Georgia Institute of Technology, Visiting Research Scholar,
Atlanta, GA, USA
|
Sept. 2018 - July 2019
|
- Advised by Prof. Dhruv Batra and Prof. Devi Parikh. Performing research in computer vision, and natural language processing.
- Worked on generating natural language navigation instructions for trajectories in Matterport3D (a photo-realistic 3D environment). Developed a hard-attention model, trained via gumbel-softmax straight through estimator. Published at CVPR 2019 (see Publications).
- Now fine-tuning instruction generation using reinforcement learning methods such as self-critical sequence training and actor-critic.
Indian Institute of Science, Research Intern,
Bengaluru, India
|
Nov. 2017 - July 2018
|
Advised by Prof. Ambedkar Dukkipati. Developed a decoder-less variant of Skip-Thought Vectors (maps natural language sentences to vectors ) which trained 10 times faster. Worked on generative adversarial networks (GANs) for semi-supervised learning. Developed models that explicitly learned to generate functions which in turn access a knowledge base (KB) for answering questions.
Software Robotics Corporation (Soroco), Platform Engineer,
Bengaluru, India
|
June 2016 - Oct. 2017
|
- Led a team which used recurrent neural networks (RNNs) to automatically find easy-to-automate business processes (for e.g. payroll management, settling vendor disputes for top e-commerce websites).
- Designed and implemented a patent-pending work queue-based distributed systems framework operational at Fortune 500 companies. The framework allowed developing efficient and fault-tolerant automation systems with minimal internal state.
Facebook, Software Engineering Intern,
Menlo Park, CA, USA
|
May 2015 - July 2015
|
Worked with the configuration management team in the Core Engineering Group. Found and fixed bottlenecks in existing architecture. Parallelized configuration checking service by splitting into master and slaves, speeding it up by 1000 times.
Fast GPU-Based Simulator for Room-to-Room dataset |
Jan. 2019
|
Optimized the Room-to-Room dataset's original simulator to only use GPU for updating simulator's state corresponding to the actions taken by the agent. The optimized simulator has 17x faster environment traversal on just a single GPU!
ICLR 2018 reproducibility challenge: Interpretable Counting for Visual Question Answering |
June 2018
|
Robot Soccer, IIT Kharagpur |
May 2013 - May 2016
|
- Led the A.I. team. Created a framework for learning robot control policies for playing soccer using Q-Learning.
- Bronze medal in FIRA 2015, South Korea.
Visual Landmark Selection for Generating Grounded and Interpretable Navigation InstructionsS. Agarwal, D. Parikh, D. Batra, P. Anderson, S. Lee |
June 2019
|
[PDF] CVPR'19 Workshop on Deep Learning for Semantic Visual Navigation
AppTechMiner: Mining Applications and Techniques from Scientific Articles, M. Singh, S. Dan, S. Agarwal, P. Goyal, A. Mukherjee |
June 2017
|
[PDF] Joint Conference on Digital Libraries (JCDL) 2017:6th International Workshop On Mining Scientific Publications