Reinforcement learning of recurrent neural network for temporal coding

Abstract

We study a reinforcement learning for temporal coding with neural network consisting of stochastic spiking neurons. In neural networks, information can be coded by characteristics of the timing of each neuronal firing, including the order of firing or the relative phase differences of firing. We derive the learning rule for this network and show that the network consisting of Hodgkin-Huxley neurons with the dynamical synaptic kinetics can learn the appropriate timing of each neuronal firing. We also investigate the system size dependence of learning efficiency.

Properties of the Internal Clock: First- and Second-Order Principles of Subjective Time

“…Transfer time would be directly proportional to the size of the oscillation vector or the number of pulses in the accumulator and the speed of transfer (i.e., baud rate). Pulse accumulation essentially functions as an up counter and memory transfer functions as a down counter, with the additional assumption that neural networks can function as look-up or conversion tables (Dali & Zemin 1993, Kimura & Hayakawa 2008.…”

A Reinforcement Learning Framework for Spiking Networks with Dynamic Synapses

“…Eligibility denotes synapses that have contributed to either a correct or false output spike. These eligible synapses can be determined either analytically as in [ 15 , 17 , 40 ] or phenomenologically as in [ 16 , 22 , 25 ]. In order to keep complexity at a minimum, the latter approach is the one adopted in the presented study.…”