My career goal is to build machines that improve exponentially 📈.
2022 - Present Vice President of AI, Halodi Robotics
2016 - 2022 Senior Research Scientist at Robotics at Google
The last decade of Machine Learning research has shown that rich data + simple learning algorithms + powerful computational resources outperforms hand-engineered systems in computer vision, translation, and NLP problems. My research focused on applying that principle to the domain of robotics.
- Co-inventor of the Gumbel-Softmax / Concrete distribution.
- Co-Lead for the Brain Moonshot team (20+ FTEs) that produced SayCan. Trained end-to-end neural networks capable of hundreds of manipulation tasks, including standing bottles upright, pick-and-place for a variety of objects, opening and closing drawers from monocular RGB images. Ran the data management work and worked on incentive alignment within the team.
- Built a shared autonomy system to open latched doors with neural networks. Featured on WIRED magazine.
- Several of my publications are featured on the Google AI blog: Semantic Grasping, QT-Opt, Grasp2Vec, RetinaGAN, BC-Z.
- Developed Tensor2Robot, a ML framework used by the robotics manipulation team and Everyday Robots (until TensorFlow 1 was deprecated).
- First author on 7 papers, coauthored 15+ more. Select works described here, and a full list of publications can be found on Google Scholar.
- Area-chair for ICML, CoRL, and reviewer for many other conferences and workshops.
2013-2015: Various internships.
I write machine learning tutorials and miscellaneous thoughts about technology on my blog.
I enjoy cooking and baking and backpacking. I am learning how to make a Tornado Omlette, without much success.
- 2012 - 2016 ScM in Computer Science (Concurrent Masters), Brown University
- 2012 - 2016 ScB in Applied Mathematics and Computer Science, Brown University