Generating coherent patterns of activity from chaotic neural networks

Area of Science:

• Computational neuroscience
• Systems neuroscience

Background:

• Neural circuits exhibit complex spontaneous and stimulus-evoked activity patterns.
• The relationship between spontaneous neural activity and functional network output remains an open question.

Purpose of the Study:

• To develop a method for modifying neural network activity from chaotic spontaneous patterns to desired functional outputs.
• To investigate the role of synaptic plasticity in shaping network dynamics.

Main Methods:

• Developed a procedure termed FORCE learning to modify synaptic strengths in model neural networks.
• Applied FORCE learning to networks with intact, unclamped feedback loops, even when exhibiting chaotic spontaneous activity.
• Trained networks to generate diverse complex output patterns, including memory-dependent transformations and controllable multi-output dynamics.

Main Results:

• Successfully transformed chaotic spontaneous neural activity into a wide range of desired complex patterns using FORCE learning.
• Demonstrated the ability to create networks capable of input-output transformations requiring memory and switching between multiple outputs.
• Generated motor patterns that closely matched human motion capture data.
• Replicated premovement activity patterns observed in motor and premotor cortex.

Conclusions:

• FORCE learning is an effective method for controlling neural network activity and generating complex functional patterns.
• Synaptic plasticity can act as a rapid and potent modulator of neural network dynamics.
• The findings suggest that rapid synaptic plasticity may play a significant role in shaping neural computations and motor control.