I am a Senior Research Scientist at Google Brain in Toronto. I am interested in developing simple and efficient machine learning algorithms that are broadly applicable across a range of application domains including natural language processing and computer vision. My personal website can be found at norouzi.github.io.
Currently, I am highlighting:
- OCD: Optimal Completion Distillation -- a training procedure for optimizing sequence to sequence models directly based on edit distance. OCD achieves the state-of-the-art performance on end-to-end speech recognition. (ICLR'19)
- Keypointnet -- an effective approach to discovery of latent 3D keypoints from 2D images via end-to-end geometric reasoning. (NeurIPS'18 Oral) [code] [web]
- MAPO: Memory Augmented Policy Optimization -- an unbiased estimator of policy gradients that incorporates off-policy trajectories without importance sampling. MAPO achieves the state-of-the-art results on semantic parsing. (NeurIPS'18 Spotlight) [code]
- Bridging the gap between value and policy based reinforcement learning via entropy regularization. [Follow-up] [code]
- RAML: Reward Augmented Maximum Likelihood -- an objective born from the marriage of conditional log-likelihood and expected return. [poster] [slides]