Integrated the trajectories generated by imitation learning (IL) into the reinforcement learning (RL) pipeline and analyzed the planning behavior of resulting models.

Implemented a memory based architecture (using RNNs) in the distributed RL framework to help with the partial observability of environments (inspired by DeepMind’s R2D2).

Collaborated with the AI Habitat team to develop a high-performance 3D simulator and an API to research autonomous navigation tasks.

Researched reinforcement learning based approaches for the PointGoal navigation task and achieved the best performance on the Habitat 2019 RGB challenge.

Implemented a Las Vegas transformation of the Markov Chain Monte Carlo estimators to improve the training of Restricted Boltzmann Machines (RBMs).

Conducted experiments on the MNIST dataset to demonstrate that the transformation achieved a higher likelihood and outperformed the state-of-the-art Contrastive Divergence method.

Implemented the automatic user provisioning feature of Google’s Identity as a Service (IDaaS)

platform.

Developed a pipeline to automate user creation and management from G Suite to third-party applications.