Supervised learning in spiking neural networks with FORCE training.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Neural Network Modeling

Background:

• Neural populations exhibit diverse behaviors.
• Machine learning offers tools to model complex neural dynamics.
• The FORCE method is a machine learning technique for training recurrent neural networks.

Purpose of the Study:

• To demonstrate the applicability of the FORCE method to spiking neural networks.
• To train spiking neural networks to perform complex behavioral tasks.
• To create biologically motivated neural network models.

Main Methods:

• Applied the FORCE method to train spiking neural networks.
• Trained networks to mimic dynamical systems.
• Trained networks for input classification and sequence storage.
• Developed biologically inspired models of songbird and hippocampal circuits.

Main Results:

• FORCE-trained networks successfully mimicked dynamical systems, classified inputs, and stored sequences.
• Biologically motivated networks reproduced complex behaviors like songbird singing and movie scene replay.
• Achieved behaviors comparable in complexity to the biological systems they were inspired by.

Conclusions:

• The FORCE method is directly applicable to spiking neural networks for creating complex behaviors.
• FORCE-trained models provide insights into neural circuit function, including responses to perturbations.
• This technique offers a powerful approach for understanding neural computation and network dynamics.