Dwait Bhatt

Experience

• 

  CSIR-CEERI

  Sept 2017 - May 2018

  Research Project Member

  Developed a Reinforcement Learning agent that learned to control the temperature and humidity of a closed chamber using devices interfaced with it via an Arduino.

• 

  National University of Singapore

  Jun 2018 - Jun 2018

  Academic Intern

  Academic Internship - Studied deep learning with a focus on CNNs starting from LeNet to ResNet, analyzing the techniques they introduced to Computer Vision, and why they worked well.

• 

  Pixxel

  Aug 2018 - Dec 2018

  Machine Learning Engineer

  Worked on semantic segmentation of satellite imagery to identify objects like roads, vehicles, buildings and water bodies as part of a project in partnership with MapMyIndia.

• 

  Samsung Electronics

  Jan 2019 - Aug 2022

  - Spearheaded the development of a patented application launch time prediction model, from POC to on-device implementation. This was used to optimize CPU boosting and make application launches 6% faster while extending battery life by 2 hours.- Saved 10 days per quarter of manual work by creating a tool to extract user-perceived launch time from app launch videos.- Authored a research paper based on this work and presented it at IEEE IJCNN 2022 held in Padua, Italy. - Developed a Reinforcement Learning agent that controlled Android kernel tunables to improve device performance in application launch and video streaming scenarios. Created various automation tools to determine the impact of Android kernel tunables on average application launch time.

  • Senior Machine Learning Engineer

    Mar 2021 - Aug 2022

  • Machine Learning Engineer

    Jul 2019 - Feb 2021

  • Intern

    Jan 2019 - Jun 2019
• 

  ISRO - Indian Space Research Organization

  Sept 2021 - Jan 2022

  Research Trainee

  Reduced the computational and monetary cost of processing satellite imagery using a Reinforcement Learning system that learned to identify image patches essential for downstream tasks like image classification and object detection.

• 

  UC San Diego

  Jan 2023 - now

  Graduate Researcher

  - Developed a sim-to-real pipeline to transfer learnt robosuite policies onto a physical xArm6 robot.- Taught robots to perform manipulation tasks from demonstrations using GAIL. - Set up VR teleoperation to control a real robot arm and capture task demonstrations. - Currently working on learning latent policies to enable transfer across robot types.Advisor: Nikolay Atanasov