Real-Time Decision-Making via Data-Driven Optimization

Prof. Bartolomeo Stellato
Princeton University
DEIB - Conference Room "Emilio Gatti" (Building 20)
On Line via Webex
June 16th, 2022
3.00 pm
Contacts:
Maria Prandini
Research Line:
Control systems
Princeton University
DEIB - Conference Room "Emilio Gatti" (Building 20)
On Line via Webex
June 16th, 2022
3.00 pm
Contacts:
Maria Prandini
Research Line:
Control systems
Sommario
On June 16th, 2022 at 3.00 pm Bartolomeo Stellato, Assistant Professor at Princeton University, will give a seminar on "Real-Time Decision-Making via Data-Driven Optimization" in DEIB Conference Room and in live streaming via Webex.
We present machine learning tools to design and efficiently compute control policies involving the real-time solution of convex optimization problems. In the first part of the talk, we introduce a machine learning approach to accelerate OSQP, a popular convex quadratic optimization solver that is numerically robust and suitable for embedded applications. However, being a first-order method, OSQP can exhibit slow convergence in case of badly scaled data. To overcome these limitations, we use reinforcement learning to train a new constraint-wise step-size update rule and reduce the average runtime by up to 30%.
In the second part of the talk, we introduce an end-to-end architecture to learn control policies by varying the parameters of the underlying decision-making problem. With this approach, which relies on recent techniques to differentiate through the solution of convex optimization problems, we can automatically learn control policies in finance, robotics, and supply chain management.
We present machine learning tools to design and efficiently compute control policies involving the real-time solution of convex optimization problems. In the first part of the talk, we introduce a machine learning approach to accelerate OSQP, a popular convex quadratic optimization solver that is numerically robust and suitable for embedded applications. However, being a first-order method, OSQP can exhibit slow convergence in case of badly scaled data. To overcome these limitations, we use reinforcement learning to train a new constraint-wise step-size update rule and reduce the average runtime by up to 30%.
In the second part of the talk, we introduce an end-to-end architecture to learn control policies by varying the parameters of the underlying decision-making problem. With this approach, which relies on recent techniques to differentiate through the solution of convex optimization problems, we can automatically learn control policies in finance, robotics, and supply chain management.
Biografia
Bartolomeo Stellato is an Assistant Professor in the Department of Operations Research and Financial Engineering at Princeton University. Previously, he was a Postdoctoral Associate at the MIT Sloan School of Management and Operations Research Center. He received his D.Phil. (Ph.D.) in Engineering Science from the University of Oxford as part of the Marie Curie EU project TEMPO. He holds a M.Sc. in Robotics, Systems and Control from ETH Zürich and a B.Sc. in Automation Engineering from Politecnico di Milano. He is the developer of OSQP, which is one of the most widely used software in mathematical optimization. Bartolomeo Stellato is the recipient of the 2021 Princeton SEAS Innovation Award in Data Science, the 2020 Best Paper Award in Mathematical Programming Computation, the 2020 INFORMS Pierskalla Best Paper Award, and the 2017 First Place Prize Paper Award in IEEE Transactions on Power Electronics. His research focuses on data-driven computational tools for mathematical optimization, machine learning, and optimal control.