Stimulus-stimulus association via reinforcement learning in spiking neural network

In this paper, we propose an algorithm that performs stimulus-stimulus association via reinforcement learning. In particular, we develop a recurrent network with dynamic properties of Izhikevich spiking neuron model and train the network to associate a stimulus pair using reward modulated spike-time dependent plasticity. The learning algorithm associates a prime stimulus, known as the predictor, with a second stimulus, known as the choice, comes after an inter-stimulus interval. The influence of the prime stimulus on the neural response after the onset of the later stimulus is then observed. A series of probe trials resemble the retrospective and prospective activities in human response processing.