Self-organizing neural architectures and cooperative learning in a multiagent environment.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Adaptive Resonance Theory (ART) is a class of self-organizing neural networks.
• Temporal-Difference (TD) methods are crucial for real-time reinforcement learning.
• Multiagent systems require robust learning and cooperation mechanisms.

Purpose of the Study:

• To investigate the cooperative learning capabilities of TD-FALCON networks in dynamic multiagent environments.
• To compare TD-FALCON performance against traditional Q-learning agents.
• To identify optimal cooperative strategies for multiagent reinforcement learning tasks.

Main Methods:

• Utilized Temporal-Difference-Fusion Architecture for Learning, Cognition, and Navigation (TD-FALCON) networks.
• Implemented minefield navigation and predator/prey pursuit tasks.
• Compared TD-FALCON with Q-learning agents trained using backpropagation and resilient-propagation (RPROP).

Main Results:

• TD-FALCON agent teams successfully adapted and performed well in multiagent navigation without explicit collaboration.
• TD-FALCON significantly outperformed traditional Q-learning agents in navigation tasks.
• A combination of compressed state representation and a hybrid reward function yielded optimal results for the predator/prey task.

Conclusions:

• TD-FALCON offers a powerful framework for cooperative learning in multiagent systems.
• TD-FALCON demonstrates superior learning efficiency and task completion rates compared to RPROP-based reinforcement learners.
• The proposed cooperative strategy enhances performance in complex multiagent reinforcement learning scenarios.