Multiagentní zpětnovazební učení

• Thesis title in Czech: Multiagentní zpětnovazební učení
• Thesis title in English: Multi-agent reinforcement learning
• Academic year of topic announcement: 2021/2022
• Thesis type: dissertation
• Thesis language: čeština
• Department: Department of Theoretical Computer Science and Mathematical Logic (32-KTIML)
• Supervisor: Mgr. Martin Pilát, Ph.D.
• Author: hidden - assigned and confirmed by the Study Dept.
• Date of registration: 20.09.2021
• Date of assignment: 20.09.2021
• Confirmed by Study dept. on: 24.09.2021

Guidelines

Reinforcement learning (RL) has recently outperformed human experts in many games such as chess, go and even some computer games. Solving these problems was considered difficult due to their nature - large action space, rewards delayed to the end of the game, and no simple expert rules for a victory. The potential of reinforcement learning (especially Deep RL) to deal successfully with all these problems raises big expectations for the future of artificial intelligence (AI).

At the same time, various important control problems such as traffic management, automated warehouses, autonomous cars, or cooperative and competitive games can be treated as multi-agent RL (MARL) problems. These share similar problems with large action spaces and delayed rewards with the single-agent ones, and similar techniques, including evolutionary algorithms and neural networks, are used to solve them.

The goal of the thesis is to explore and improve multi-agent reinforcement learning techniques considering both simple toy problems and more practical ones like those mentioned above and others, such as inter-agent communication learning or multi-agent path finding. One of the goals is also to compare the various techniques used to solve MARL problems, find their strong and weak points, and potentially integrate them to create more powerful ones.

References

[1] Sutton, Richard S., and Andrew G. Barto. "Reinforcement learning: An introduction." MIT press, 2018. ISBN: 978-0-26-203924-6
[2] Ian Goodfellow and Yoshua Bengio and Aaron Courville. "Deep Learning." MIT Press, 2016. ISBN: 978-0-26-203561-3
[3] Foerster, Jakob, Gregory Farquhar, Triantafyllos Afouras, Nantas Nardelli, and Shimon Whiteson. "Counterfactual Multi-Agent Policy Gradients." ArXiv:1705.08926, 2017. http://arxiv.org/abs/1705.08926
[4] Hernandez-Leal, Pablo, Bilal Kartal, and Matthew E. Taylor. "Is Multiagent Deep Reinforcement Learning the Answer or the Question? A Brief Survey." ArXiv:1810.05587, 2018. http://arxiv.org/abs/1810.05587
[5] Mnih, Volodymyr, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, et al. "Human-Level Control through Deep Reinforcement Learning." Nature 518, no. 7540, 2015: 529–33. DOI: 10.1038/nature14236