Distributed Optimization | Artificial Intelligence Institute for Advances in Optimization

The sheer size and complexity of many modern multi-agent, networked systems mean that often data collection, computing, and decision making do not all happen at a central location. AI algorithms working in real-time and in a decentralized manner will become of utmost importance as advances in sensor technology and wireless communications make it possible to collect and transmit tremendous amounts of data between many locations.

This thrust extends recent results in centralized algorithms for non-convex continuous optimization to the decentralized setting, which has remained elusive. It also develops novel approaches for Distributed Constraint Optimization Problems (DCOP) which play a central role in multi-agent reasoning. In addition, It studies how agents can accelerate their learning even when they learn policies for different objectives by exploiting discovered commonalities in their environments.

Publications

Related work

Finite-time analysis of decentralized stochastic approximation with applications in multi-agent and multi-task learning. S. Zeng, T. Doan, and J. Romberg. arxiv:2010.15088, October 2020.
Communication-efficient algorithms for decentralized and stochastic optimization. G. Lan, S. Lee, and Y. Zhou. Mathematical programming, 180(1):237–284, 2020.
Lifelong multi-agent path finding in large-scale warehouses. J. Li, A. Tinka, S. Kiesel, J. Durham, S. Kumar, and S. Koenig. In Proceedings of the International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS), pages 1898–1900, 2020.
Fast convergence rates of distributed subgradient methods with adaptive quantization. T. Doan, S. T. Maguluri, and J. Romberg. Accepted to IEEE Trans. Auto. Control (arxiv:1810.13245), November 2019.
Task and path planning for multi-agent pickup and delivery. M. Liu, H. Ma, J. Li, and S. Koenig. In Proceedings of the International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS), pages 1152–1160, 2019.
Primal: Pathfinding via reinforcement and imitation multi-agent learning. G. Sartoretti, J. Kerr, Y. Shi, G. Wagner, S. Kumar, S. Koenig, and H. Choset. IEEE Robotics and Automation Letters, 4(3):2378– 2385, 2019.
Solving multi-agent constraint optimization problems on the constraint composite graph. Ferdinando Fioretto, Hong Xu, Sven Koenig, and T. K. Satish Kumar.Proceedings of the Twenty-First International Conference on Principles and Practice of Multi-Agent Systems, 2018.