Bayesian neural networks in the multi-agent environment

Authors: Podmaryov M.S.
Published in issue: #3(92)/2024
DOI:
Category: Informatics, Computer Engineering and Control | Chapter: Information Technology. Computer techologies. Theory of computers and systems
Keywords: multi-agent environment, machine learning, multi-agent reward learning, neural networks, Bayesian neural networks, Bayesian Actor-Critic algorithm
Published: 25.07.2024

The paper considers behavioral features of the Bayesian neural networks in the multi-agent environment. It presents reward learning, multi-agent reward learning, and the Bayesian Actor-Critic behavior in the multi-agent environment. Learning using the Bayesian neural networks is analyzed. The Bayesian neural network architecture and hierarchy are demonstrated. The paper describes agent components and their interaction with the testing environment. The Mujoco environment is selected in testing. The environment features and aspects are highlighted. Tests were conducted to check behavior of the agents controlled by the Bayesian neural network using the Actor-Critic algorithm. Test results are presented, they prove the algorithm efficiency, achieving successful learning, and interaction between the agents and the environment. The work performed is acknowledging that the Bayesian neural networks and the Bayesian Actor-Critic algorithm are capable of learning and achieving set results in the multi-agent environment.

References

[1] Lapan M. Deep Reinforcement Learning. AlphaGo and Other Technologies. Saint Petersburg, Piter, 2020, 496 p.(In Russ.).

[2] Sutton R.S., Barto E.J. Reinforcement Learning. Moscow, DMK Press, 2020, 552 p.(In Russ.).

[3] Alfimtsev A.N. Multi-Agent Reinforcement Learning. Moscow, Bauman Moscow State Technical University, 2021, 222 p.(In Russ.).

[4] Konda V. Actor-Critic Algorithms. Available at: https://dspace.mit.edu/bitstream/handle/1721.1/8120/51552606-MIT.pdf?sequence=2 (accessed April 15, 2024).

[5] Downey A.B. Bayesian Models. Moscow, DMK Press, 2019, 184 p.(In Russ.).

[6] Qin Yang, Ramviyas Parasuramana. A Strategy-Oriented Bayesian Soft Actor-Critic Model. The 14th International Conference on Ambient Systems, Networks and Technologies (ANT), 2023, Leuven, Belgium. Available at: https://arxiv.org/pdf/2303.04193.pdf (accessed March 7, 2023).