CoBeL-RL: A neuroscience-oriented simulation framework for complex behavior and learning.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Reinforcement learning (RL) is increasingly used to model animal behavior and neuronal activity.
• Computational neuroscience software is fragmented, hindering method development and result comparison.
• Existing machine learning tools are not well-suited for neuroscience research requirements.

Purpose of the Study:

• Introduce CoBeL-RL, a closed-loop simulator for complex behavior and learning.
• Provide a neuroscience-oriented framework for efficient simulation setup and execution.
• Address the need for integrated and comparable software in computational neuroscience.

Main Methods:

• Developed CoBeL-RL, a simulator based on RL and deep neural networks.
• Integrated virtual environments (T-maze, Morris water maze) with varying abstraction levels.
• Included multiple RL algorithms (e.g., Dyna-Q, deep Q-network) and analysis tools.

Main Results:

• CoBeL-RL offers a neuroscience-oriented framework for efficient simulations.
• The simulator supports diverse virtual environments and RL algorithms.
• Provides tools for behavior and unit activity analysis with closed-loop control.

Conclusions:

• CoBeL-RL bridges a critical gap in computational neuroscience software.
• Facilitates the development, integration, and comparison of RL-based research.
• Enhances the study of learning and neuronal representations.