- Poster presentation
- Open Access
Controlling neuronal fluctuations for directed exploration during reinforcement learning
BMC Neurosciencevolume 10, Article number: P138 (2009)
Neuronal and synaptic fluctuations have both been proposed to underly reward controlled learning [1, 2] and have been used to explain song learning in songbird area RA . The songbird area LMAN provides perturbations to area RA that are necessary for learning , suggesting that LMAN might target specific subsets of RA neurons and control the corresponding noise level for directed experimentation. Here we explore this hypothesis by investigating algorithms for controlling the amount of noise in order to yield efficient reinforcement learning in large networks. Our research is guided by previous work on exploration for learning which exploits information gain . We find that noise control can strongly increase learning efficiency thereby attenuating the curse of dimensionality. Our results suggest that area LMAN controls experimentation by targeted control and injection of noise into RA, which might have testable implications also for learning in other motor pathways.
Xie X, Seung HS: Learning in neural networks by reinforcement of irregular spiking. Physical Review E 69. 2004, 041909-10.1103/PhysRevE.69.041909.
Seung HS: Learning in spiking neural networks by reinforcement of stochastic synaptic transmission. Neuron. 2003, 40: 1063-1073. 10.1016/S0896-6273(03)00761-X.
Fiete I, Fee M, Seung HS: Model of birdsong learning based on gradient estimation by dynamic perturbation of neural conductances. Journal of Neurophysiology. 2007, 98: 2038-2057. 10.1152/jn.01311.2006.
Ölveczky B, Andalman A, Fee M: Vocal experimentation in the juvenile songbird requires a basal ganglia circuit. PLoS Biol. 2005, 3: e153-10.1371/journal.pbio.0030153.
Si B, Pawelzik K: Robot exploration by subjectively maximizing objective information gain. Robotics and Biometrics IEEE International Conference. 2004, 930-935.