End to End Learning and Optimization

This methodology thrust investigates three research avenues:

Optimization Proxies: the idea of replacing a time-consuming optimization model by a machine-learning proxy that can be used in real time and/or in computational demanding applications (e.g., large-scale Monte-Carlo simulations). The challenge is to learn such proxies for large-scale optimization problems that include physical, engineering, and operational constraints. The proxies can also be improved by reinforcement learning when they are part of as sequential decision process.
End-to-End Learning: the idea of integration of optimization layers as parts of the deep-learning pipeline. The challenge is to define combinatorial layers capturing specific families of combinatorial optimization tasks that are amenable to useful differentiable optimization surrogates.
Learning to optimize: the idea of replacing components of optimization models by machine-learning models.

Here are a teaser and a longer presentation about optimization proxies with applications in energy, supply chains, and mobility.

<br />

The Institute is pioneering a number of new methodological directions in this space:

The Just-In-Time Learning methodology which trains machine-learning models in real time to account for changes in operational environments.
Self-supervised learning for constrained optimization which learns both primal and dual values in an augmented Lagrangian framework.
The combination of reinforcement learning and optimization proxies inside a model-predictive framework to capture both short-term and long-term rewards.
The combination of confidence-aware graph-neural networks and fast feasibility restoration to guarantee feasibility.
The compact optimization learning methodology where the learning takes place in a subspace of principal components of the outputs.
Warm-starting ADMM Algorithms with optimal primal and dual estimates.
Two-stage learning for scheduling problems featuring a combination of machine allocation and machine scheduling.
Online and Offline Learning for Contextual Stochastic Stochastic Optimization

These methodologies have been applied to real-case studies in power systems and transportation systems.

UCLA IPAM Presentation on Online and Offine Learning for Contextual Stochastic Learning

<br />

Publications

Funded by the Institute

Learning Regionally Decentralized AC Optimal Power Flows with ADMM, Terrence Mak, Minas Chatzos, Mathieu Tanneau, and Pascal Van Hentenryck. IEEE Transactions on Smart Grids, 14(6), 4863-4876. November 2023.
End-to-End Feasible Optimization Proxies for Large-Scale Economic Dispatch. Wenbo Chen, Mathieu Tanneau, and Pascal Van Hentenryck. IEEE Transactions on Power Systems (to appear).
Compact Optimization Learning for AC Optimal Power Flow, Seonho Park, Wenbo Chen, Terrence W.K. Mak, and Pascal Van Hentenryck, IEEE Transactions on Power Systems (to appear).
Constraint Programming to Improve Hub Utilization in Autonomous Transfer Hub Networks. Chungjae Lee, Wirattawut Boonbandansook, Vahid Eghbal Akhlaghi, Kevin Dalmeijer, and Pascal Van Hentenryck. The 29th International Conference on Principles and Practice of Constraint Programming (CP-2023), Toronto, Canada, August 27 - 31, 2023.
Load Embeddings for Scalable AC-OPF Learning. Terrence W.K. Mak, Ferdinando Fioretto, and Pascal Van Hentenryck. 2023 IEEE Power & Energy Society (PES) General Meeting, Orlando, Florida, July 2023.
Confidence-Aware Graph Neural Networks for Learning Reliability Assessment Commitments. Seonho Park, Wenbo Chen, Dahye Han, Mathieu Tanneau, and Pascal Van Hentenryck. IEEE Transactions on Power Systems (to appear).
Optimization-based Learning for Dynamic Load Planning in Trucking Service Networks. Ritesh Ojha, Wenbo Chen, Hanyu Zhang, Reem Khir, Alan Erera, Pascal Van Hentenryck. ArXiv:2307.04050v1, July 8, 2023.
Self-Supervised Primal-Dual Learning for Constrained Optimization. Seonho Park and Pascal Van Hentenryck. In the Thirty-Seven AAAI Conference on Artificial Intelligence (AAAI-23), February 2023.
Learning Optimization Proxies for Large-Scale Security-Constrained Economic Dispatch. Wenbo Chen, Seonho Park, Mathieu Tanneau, and Pascal Van Hentenryck. Electric Power Systems Research, Volume 213, December 2022, 108566.
Learning Optimization Proxies for Large-Scale Security-Constrained Economic Dispatch. Wenbo Chen, Seonho Park, Mathieu Tanneau, and Pascal Van Hentenryck. In the Proceedings of the 22nd Power Systems Computation Conference (PSCC), June 27 – July 1, 2022 in Porto, Portugal.
End-to-end Learning for Fair Ranking Systems. James Kotary, Ferdinando Fioretto, Pascal Van Hentenryck, and Ziwei Zhu. In the ACM Web Conference (WWW-2022), Lyon, France, April 2022.
Risk-Aware Control and Optimization for High-Renewable Power Grids. Neil Barry, Minas Chatzos, Wenbo Chen, Dahye Han, Chaofan Huang, Roshan Joseph, Michael Klamkin, Seonho Park, Mathieu Tanneau, Pascal Van Hentenryck, Shangkun Wang, Hanyu Zhang, Haoruo Zhao. arXiv:2204.00950, April 2022.
End-to-End Learning for Fair Ranking Systems. James Kotary, Ferdinando Fioretto, Pascal Van Hentenryck, and Ziwei Zhu. In the ACM Web Conference (WWW-2022), Lyon, France, April 2022.

Related work

Reinforcement Learning from Optimization Proxy for Ride-Hailing Vehicle Relocation. Enpeng Yuan, Wenbo Chen, and Pascal Van Hentenryck. Journal of Artificial Intelligence Research (JAIR), 75, 985-1002, November, 2022.
Spatial Network Decomposition for Fast and Scalable AC-OPF Learning.Minas Chatzos, Terrence Mak, and Pascal Van Hentenryck. IEEE Transactions on Power Systems, 37(4), 2601--2612, July 2022.
Spatial Network Decomposition for Fast and Scalable AC-OPF Learning. Minas Chatzos, Terrence Mak, and Pascal Van Hentenryck. IEEE Transactions on Power Systems (Early access: 10.1109/TPWRS.2021.3124726).
Machine Learning for Optimal Power Flows. Pascal Van Hentenryck. INFORMS TutORials 2021.
End- to-End Constrained Optimization Learning: A Survey. James Kotary, Ferdinando Fioretto, Pascal Van Hentenryck, and Bryan Wilder. In the 30th International Joint Conference on Artificial Intelligence (IJCAI-21), Montreal, Canada, August, 2021.
Combining Deep Learning and Optimization for Preventive Security-Constrained DC Optimal PowerFlow.Alexandre Velloso and Pascal Van Hentenryck. IEEE Transactions on Power Systems, 36(4), July 2021.
Real-time dispatching of large-scale ride-sharing systems: Integrating optimization, machine learning, and model predictive control. Connor Riley, Pascal Van Hentenryck, and Enpeng YuanIn Christian Bessiere, editor, Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI 2020, pages 4417–4423. ijcai.org, 2020.
Lagrangian Duality for Constrained Deep Learning. Ferdinando Fioretto, Pascal Van Hentenryck, Terrence WK Mak, Cuong Tran, Federico Baldo, and Michele Lombardi. In Proceedings of 2020 European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, June 2020.
Mipaal: Mixed integer program as a layer. Aaron Ferber, Bryan Wilder, Bistra Dilkina, and Milind Tambe. In AAAI Conference on Artificial Intelligence, pages 1504–1511, 2020.
Predicting AC optimal power flows: Combining deep learning and lagrangian dual methods. Ferdinando Fioretto, Terrence W.K. Mak, and Pascal Van Hentenryck. Proceedings of the AAAI Conference on Artificial Intelligence, 34(01):630–637, 2020.
Real-time dispatching of large-scale ride-sharing systems: Integrating optimization, machine learning, and model predictive control. Connor Riley, Pascal Van Hentenryck, and Enpeng Yuan. In Christian Bessiere, editor, Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI 2020, pages 4417–4423.
End-to-end game-focused learning of adversary behavior in security games. Andrew Perrault, Bryan Wilder, Eric Ewing, Aditya Mate, Bistra Dilkina, and Milind Tambe. In AAAI Conference on Artificial Intelligence, pages 1378–1386, 2020.
Melding the Data-Decisions Pipeline: Decision-Focused Learning for Combinatorial Optimization. B. Wilder, B. Dilkina, and M. Tambe. In AAAI Conference on Artificial Intelligence, 2019.
Bryan Wilder, Eric Ewing, Bistra Dilkina, and Milind Tambe. End to end learning and optimization on graphs. In Advances in Neural Information Processing Systems, pages 4672–4683, 2019.