Temporal Sequential Wave Neural Network for Solving the Optimal Cognitive Subgraph Query Problem

The Optimal Cognitive Subgraph query problem (MCSQR) aims to identify the subgraph structure that achieves the highest global recognition in a static social network. In this paper, we propose the Temporal Wave Neural Network (TSWNN) framework, which, differs from traditional neural networks, does not require any training. All neurons in TSWNN are computed in parallel and consist of input, wave receiver, neuron state memory, wave generator, wave transmitter, and output. The fundamental idea behind TSWNN is based on parallel wave transmission where each neuron can receive waves sent by all precursor neurons before activation occurs. Each neuron transmits its path and knowledge to its successor through waves while objective neurons calculate final recognition based on received waves and output optimal solutions. Evaluation using four public datasets shows that TSWNN outperforms A*, Dijkstra, Label, and TDNN.