RM-SORN: a reward-modulated self-organizing recurrent neural network

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Neural plasticity is crucial for learning and memory.
• Reward-based learning explains how the brain adapts to achieve goals.
• Understanding these mechanisms can shed light on brain capabilities.

Purpose of the Study:

• To define a neural network model integrating Intrinsic Plasticity (IP) and reward-modulated Spike-Timing-Dependent Plasticity (STDP).
• To evaluate the model's performance on diverse cognitive tasks.
• To explore the role of interacting plasticity rules in reward-modulated learning and self-organization.

Main Methods:

• Developed a neural network model incorporating IP and reward-modulated STDP.
• Tested the model on prediction, recall, non-linear computation, pattern recognition, and sequence generation tasks.
• Utilized biologically motivated plasticity rules and reward strategies.

Main Results:

• The model achieved performance comparable to supervised learning networks.
• Demonstrated effective learning through the interaction of IP and STDP.
• Showcased the potential of reward-modulated self-organization in neural networks.

Conclusions:

• The interaction of multiple plasticity rules is vital for reward-modulated learning.
• Reward-modulated self-organization may explain complex brain functions.
• This model offers a biologically plausible approach to artificial learning.