Simulating Artificial Neural Network Using Hierarchical Coloured Petri Nets
Chutiakrn Jitmit, Wiwat Vatanawood · 2021
Coloured Petri nets (CPNs) have been practically exploited to represent the high-level model of the wide ranges of applications. These particular CPN models would simulate the algorithms and provide the state space analysis and verification of the target applications. To enhance the capabilities of a coloured Petri net to perform the simulation of artificial intelligence features, this paper proposes the systematic scheme to convert the neurons and their interconnections in an artificial neural network (ANN), especially backpropagation neural network, into a module of CPN subset with ports and sockets. This resulting CPN module would be seen as a black box to perform the ANN's functionalities and eventually glued together to the other CPN modules. A set of rules are proposed to guide the converting of a backpropagation ANN, consisting of the multi-layers of neurons, the interconnecting weights, the bias constants, and the step function, into a hierarchical coloured Petri net. The resulting hierarchical coloured Petri net is correctly simulated and verified using CPN tool.