Mechanisms of reward-modulated STDP and winner-take-all in bayesian spiking decision-making circuit

Abstract

Decision making, as one of the most essential functions of the human brain, is the key neural process from sensory stimuli to neuropsychological choices till actions. More recently, numerous growing neurophysiology and neuroscience experimental evidence has indicated that the human brain performs near-optimal Bayesian inference in various tasks, such as perception, learning and decision making. In order to further understand the computational mechanism of decision-making circuit, particularly from the perspective of biological plausibility and interpretability, this paper proposes a novel brain-inspired decision-making circuit based on spiking neural networks for perceptual decision-making tasks. The proposed model employs a winner-take-all (WTA) mechanism and reward-modulated spike-timing-dependent plasticity (STDP) related with Bayesian computation to simulate the neural representation of decision-making. Experiments in the random-dot motion discrimination task demonstrate that the proposed spiking decision-making circuit exhibits WTA property and has a better performance compared with unsupervised STDP.