The role of chaotic resonance in cerebellar learning.

According to the cerebellar learning hypothesis, the inferior olive neurons, despite their low firing rates, are thought to transmit high-fidelity error signals to the cerebellar cortex. "Chaotic resonance", via moderate electrical coupling between inferior olive neurons, has been proposed to realize efficient transmission of the error signal by desynchronizing spiking. Here, we first show that chaotic resonance is a robust phenomenon, as it does not depend upon the details of the inferior olive neuronal model. Second, we show that chaotic resonance enhances learning of a neural controller for fast arm movements. Furthermore, when both coupling and noise are considered simultaneously, we found that chaotic resonance widens the range of noise intensity within which efficient learning can be realized. We suggest that, from an energetic viewpoint, the spiking activity induced by chaos can be more economical than that induced by noise.