Research Engineer, Google DeepMind
navoditasharma[at]google[dot]com
navoditasharma16[at]gmail[dot]com


Research Interests
Reinforcement Learning
Offline RL
Generative Models
Privacy

Bio

I am a Research Engineer at Google DeepMind. My research interests lie primarily in the area of Reinforcement Learning and its applications.

My latest research involved developing differentially private algorithms for offline reinforcement learning, in collaboration with Dr. Alekh Agarwal, Dr. Abhradeep Guha Thakurta and Dr. Christoph Dann.

Currently, I’m exploring how reinforcement learning can facilitate continual learning in LLM agents. I’m also working on using RL techniques for generating synthetic data to train LLMs.

Additionally, I’m involved in a project aimed at automating the discovery of interpretable policies from neuroscience data. In the past, I have worked on developing solutions for learning from aggregated data to preserve data privacy, mentored by Dr. Aravindan Raghuveer, Dr. Rishi Saket and Dr. Karthikeyan Shanmugan.

Before this role, I was a Research Intern at Google Research India. I worked with Dr. Aravindan Raghuveer and Prof. Balaraman Ravindran on Imitation Learning in the absence of environmental access. Prior to this, I interned at Viterbi School of Engineering USC where, under the guidance of Prof. Xiang Ren, I worked on Time Series prediction using Knowledge Graphs.

I completed my Dual Degree (Bachelors + Masters) in Computer Science and Engineering from Indian Institute of Technology, Madras in 2021. My Dual Degree Thesis, under the guidance of Prof. Balaraman Ravindran, focused on Behavioral Cloning for multi-agent Traffic Signal Control.

Updates

  • [January 2024] Our paper Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation got accepted to ICLR 2024.
  • [December 2023] Our paper Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation got accepted to Regulatable ML Workshop at NeurIPS 2023 as an Oral Presentation. I presented this work at NeurIPS 2023 in New Orleans.
  • [July 2023] Attended Conference on Learning Theory (COLT), 2023 in Bengaluru.
  • [July 2022] Joined Google Research India as a Research Software Engineer in the Advertising Sciences Team.
  • [June 2021] Completed my Dual Degree (Bachelors + Masters) in Computer Science and Engineering from IIT Madras.
  • [April 2021] Joined Google Research India as a Research Intern in the Advertising Sciences Team.
  • [January 2021] Attended International Joint Conference on Artificial Intelligence (IJCAI), 2020 virtually. Presented my paper as a poster.
  • [April 2020] Our paper Temporal attribute prediction via joint modeling of multi-relational structure evolution got accepted to IJCAI 2020 as a poster.
  • [May 2019] Joined Viterbi School of Engineering, USC as a Research Intern. Was one of the 15 students selected from India through the IUSSTF-Viterbi program.

Publications & Preprints

Preserving Expert-Level Privacy in Offline Reinforcement Learning

Navodita Sharma, Vishnu Vinod, Abhradeep Thakurta, Alekh Agarwal, Borja Balle, Christoph Dann, Aravindan Raghuveer
arXiv preprint arXiv:2411.13598 (Under review at AISTATS 2025) Paper

Learning from Label Proportions and Covariate-shifted Instances

Sagalpreet Singh, Navodita Sharma, Shreyas Havaldar, Rishi Saket, Aravindan Raghuveer
arXiv preprint arXiv:2411.12334 (Under review at AISTATS 2025) Paper

Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation

Shreyas Havaldar*, Navodita Sharma*, Shubhi Sareen, Karthikeyan Shanmugam, Aravindan Raghuveer
ICLR 2024, Oral @ Regulatable ML Workshop, NeurIPS 2023 Paper

Temporal attribute prediction via joint modeling of multi-relational structure evolution

Sankalp Garg*, Navodita Sharma*, Woojeong Jin, Xiang Ren
IJCAI 2020 Paper