An Intelligent And Transparent Inference: Spiking Neural Network For Causal Reasoning

Abstract

In light of mining large amounts of data, artificial intelligence (AI) has learned a very strong correlation between objects. However, its limitations lie in that it can’t summarize the causality between objects like human beings and it forms the blind box association mechanism. In this paper, we address these limitations and make the contributions: We propose an implementation of causal reasoning based on spiking neural network (SNN), which simulates causality using information processing with spiking activities. And the spike-timing-dependent plastic rules (STDP) is utilized as a method of causal reasoning, which is based on the topological structure of causal graph and can make the process visible. Through experiments, our model completes the inference of causal ladder proposed by Judea Pearl, and experiments prove that it can complete more complex causal reasoning under the condition of integrating multiple causality.