Research on parallel computing of the olfactory neural network based on multithreading.

To improve the computational efficiency of olfactory neural network, this paper proposes a multithreading-based parallel computing method. Firstly, focusing on the olfactory neural network and its neuronal equations, this paper analyzes and compares the computational efficiency and recognition performance of the fourth-order Runge-Kutta method (RK4) and other ODE solvers in the olfactory neural network. This establishes the foundation for subsequent ODE solver selection. Then, while maintaining recognition accuracy, this paper implements parallel computing in olfactory neural networks through data parallelism based on multithreading, aiming to improve their computational efficiency. Experimental results indicate that when balancing computational speed and recognition accuracy, the forward Euler method emerges as the most suitable ODE solver for the neuronal equations of the olfactory neural network. Its average computation time is 7205 s, and the average recognition accuracy is 98.19%. The multithreading-based data parallel computing method proposed in this paper can effectively improve the computational efficiency of the olfactory neural network. When the number of threads is 12, the olfactory neural network achieves relatively the best experimental results: the average speedup ratio S is 7.9082, the parallel efficiency E is 0.6590, the average recognition accuracy is 98.30%, and the computation time is 1,359 s. Notably, when the number of threads is 5, the olfactory neural network reaches the highest average recognition accuracy of 98.77% on the epileptic EEG dataset.