It is widely accepted that the complex dynamics characteristic of recurrentneural circuits contributes in a fundamental manner to brain function. Progresshas been slow in understanding and exploiting the computational power ofrecurrent dynamics for two main reasons: nonlinear recurrent networks oftenexhibit chaotic behavior and most known learning rules do not work in robustfashion in recurrent networks. Here we address both these problems bydemonstrating how random recurrent networks (RRN) that initially exhibitchaotic dynamics can be tuned through a supervised learning rule to generatelocally stable neural patterns of activity that are both complex and robust tonoise. The outcome is a novel neural network regime that exhibits bothtransiently stable and chaotic trajectories. We further show that the recurrentlearning rule dramatically increases the ability of RRNs to generate complexspatiotemporal motor patterns, and accounts for recent experimental datashowing a decrease in neural variability in response to stimulus onset.
展开▼
