Research Engineer
- Deep learning research for Google Brain.
- Led research on modular neural architectures for program synthesis with deep reinforcement learning (RL) dynamic routing, deep RL agents with backtracking actions, and deep RL for program synthesis with open sourced.
- Implemented distributed reinforcement learning using Impala, A3C, and asynchronous deep Q-learning at scales of 10-200 CPU workers per instance with GPU/TPU gradient aggregator, and 10-400 instances per hyperparameter.
- Implemented and trained LSTM and DC-GAN melody generation models and core infrastructure for the Magenta group (art and music generation).
- Implemented training enhancements from the paper "Reward Augmented Maximum Likelihood" in Google's production neural machine translation model.