Naive Bayes classifier based on memristor nonlinear conductance

In this work,a naive Bayes classifier (NBC) based on memristor nonlinear conductance modulation is proposed, which not only can effectively avoid the influence of memristor nonlinearity and asymmetry on the network performance, but also enable on-chip training and inference completely on the memristive array. The performance of this classifier is evaluated by MNIST dataset classification, with highest recognition rate reaching 84.43%. In addition, the influence of other non-ideal factors of the memristor on the classification performance is analyzed, and a method to improve the classifier through pruning processing is proposed. The simulation proves that the improved selection Bayesian classifier (SBC) has a higher tolerance to the non-ideal factors of the memristor than the NBC.