Shane T. Barratt
 |
Shane T. Barratt |
About me
I am the CEO of Optimal Intellect, where we build Moreau, a GPU-native differentiable convex optimization solver for AI. Moreau integrates directly into PyTorch and JAX pipelines, enabling neural networks to make decisions that satisfy hard constraints while remaining end-to-end trainable. We apply it to robotics, finance, energy systems, and more.
My research focuses on convex optimization and its applications to machine learning and control. I received the Ph.D. in electrical engineering from Stanford University in 2021 (advised by Stephen Boyd), the M.S. from Stanford in 2019, and the B.S. in EECS from UC Berkeley in 2017. I am a co-creator of cvxpylayers, which brought differentiable convex optimization layers to deep learning.
I also founded a proprietary trading firm and am an active angel investor.
Previously, I have worked at Blackrock AI Labs, Lyft Level 5, Google (Skybox Imaging/Terra Bella), Qualcomm-Atheros, and SoRoCo.
Papers
For papers listed by citation count, see Google Scholar.
2022
- Stochastic control with affine dynamics and extended quadratic costs (code)
S. Barratt and S. Boyd. IEEE Transactions on Automatic Control.
2021
- Learning convex optimization models (code)
A. Agrawal, S. Barratt, and S. Boyd. IEEE/CAA Journal of Automatica Sinica.
- Fitting feature-dependent Markov chains (code)
S. Barratt and S. Boyd. Manuscript.
- A distributed method for fitting Laplacian regularized stratified models (code, talk)
J. Tuck, S. Barratt, and S. Boyd. Journal of Machine Learning Research.
- Optimal representative sample weighting (code)
S. Barratt, G. Angeris, and S. Boyd. Statistics and Computing.
- Automatic repair of convex optimization problems (code)
S. Barratt, G. Angeris, and S. Boyd. Optimization and Engineering.
- Covariance prediction via convex optimization (code)
S. Barratt and S. Boyd. Manuscript.
- Portfolio construction using stratified models (code)
J. Tuck, S. Barratt, and S. Boyd. Machine Learning in Financial Markets: A Guide to Contemporary Practice.
- Convex optimization and implicit differentiation methods for control and estimation
S. Barratt. PhD Thesis.
- Least squares auto-tuning (code)
S. Barratt and S. Boyd. Engineering Optimization.
2020
- Minimizing a sum of clipped convex functions (code)
S. Barratt, G. Angeris, and S. Boyd. Optimization Letters.
- Convex optimization over risk-neutral probabilities (code)
S. Barratt, J. Tuck, and S. Boyd. Manuscript.
- Fitting a linear control policy to demonstrations with a Kalman constraint (code)
M. Palan, S. Barratt, A. McCauley, D. Sadigh, V. Sindhwani, and S. Boyd. Proceedings of Machine Learning Research.
- Multi-period liability clearing via convex optimal control (code)
S. Barratt and S. Boyd. Manuscript.
- Low rank forecasting (code)
S. Barratt, Y. Dong, and S. Boyd. Manuscript.
- Embedded convex optimization for control (video, code)
S. Boyd, A. Agrawal, and S. Barratt. Plenary lecture, IEEE Conference on Decision and Control.
- Learning convex optimization control policies (code)
A. Agrawal, S. Barratt, S. Boyd, and B. Stellato. Proceedings of Machine Learning Research.
2019
- Fitting a Kalman smoother to data (code)
S. Barratt and S. Boyd. Proceedings of the American Control Conference.
- Differentiable convex optimization layers (code, poster)
A. Agrawal, B. Amos, S. Barratt, S. Boyd, S. Diamond, and J. Zico Kolter. Advances in Neural Information Processing Systems.
- Learning probabilistic trajectory models of aircraft in terminal airspace from position data (code)
S. Barratt, M. Kochenderfer, and S. Boyd. IEEE Transactions on Intelligent Transportation Systems.
- Differentiating through a cone program (code)
A. Agrawal, S. Barratt, S. Boyd, E. Busseti, and W. Moursi. Journal of Applied and Numerical Optimization.
2018
- Improved training with curriculum GANs
R. Sharma, S. Barratt, S. Ermon, and V. Pande. Manuscript.
- Systems and methods for discovering automatable tasks
Y. Kim, A. Qadir, A. Narayanaswamy, R. Murty, S. Barratt, and G. Nychis. US Patent.
- Optimizing for generalization in machine learning with cross-validation gradients
S. Barratt and R. Sharma. Manuscript.
- On the differentiability of the solution to convex optimization problems
S. Barratt. Manuscript.
- A note on the inception score
S. Barratt and R. Sharma. ICML Workshop on Theoretical Foundations and Applications of Deep Generative Models.
- Cooperative multi-agent reinforcement learning for low-level wireless communication
C. de Vrieze, S. Barratt, D. Tsai, and A. Sahai. Manuscript.
- Direct model predictive control
S. Barratt. ICML Workshop on Planning and Learning.
2017
- InterpNET: neural introspection for interpretable deep learning
S. Barratt. Neurips Interpretable ML Symposium.
- Active robotic mapping through deep reinforcement learning
S. Barratt. Manuscript.
2015
- A non-rigid point and normal registration algorithm with applications to learning from demonstrations
A. Lee, M. Goldstein, S. Barratt, and P. Abbeel. International Conference on Robotics and Automation.