Research on Neural Network Hyper-Parameters Optimization Based on Firefly Algorithm

Neural networks have demonstrated numerous significant advantages, including the ability to learn quickly, convergence, self-learning, and adaptability. Specifically, neural networks can effectively avoid the problem of local minima. However, it is a challenge to determine the hyperparameters of hidden neurons in neural networks, which often limits the application and optimization of neural networks. To overcome this problem, the firefly algorithm was introduced to optimize the center position and width of the hidden layer neurons in the neural network, as well as the connection weights between the hidden layer and the output layer neurons. This approach aims to improve the performance and adaptability of neural network models and provide more efficient tools for solving complex problems. Additionally, we evaluate our method in image classification task and compare our improved model with un-improved model. From extensive experiment analysis, we can observe that our optimized mechanism can significantly enhance the classification accuracy and training speed.