Development of an Intelligent System Based on Deep Neural Network Models with Advanced Algorithms for Hyper-parameter Tuning and Weight Updates Across Diverse Datasets

This work seeks to understand how deep neural networks can be improved in terms of hyperparameters and weights on different datasets with the help of intelligent system.Regarding hyperparameters and weights optimization, the paper employs Convolutional Neural Networks (CNN) and Multi-Layer Perceptron (MLP).First, the paper compares the accuracy of the convolutional neural network and multilayer perceptron using Adam and RMS prop as optimizers.Next, propose to use the CNN model fine-tuning with Adam and RMSprop optimization algorithms, but instead of setting fixed hyperparameters, the paper uses Simulated Annealing (SA) for optimization and Differential Evolution (DE) for weight updates.Further, it examines the accuracy of the CNN models trained by the Eagle Strategy-Based Optimization (ESBO) on the MNIST database when the Differential Evolution algorithm updates the weights.This approach is then used on the CIFAR dataset.In the proposed approach, all these steps are included in detail.Apart from these image datasets, the optimization strategies applied in this work include the electric load diagrams of the years 2011-2014, air quality, and cityscapes.In terms of performance, this work also measures with 'conventional' metrics, with the MNIST and CIFAR-10 benchmarks scoring 99.5% and 89.1%, respectively, 1-2% better than prior methods.The latter approach bears higher computational costs, but this is warranted by higher accuracy of predictions and a better generalization of the model.Forecast of electricity load was verified from the time series data which also supported the proposed methods with an accuracy of 96.3%.Therefore, the results of the impact assessment indicate the relative effectiveness of different optimization algorithms and amass irrefutable evidence of the effectiveness of the proposed methods and approaches.In the study of the current sample, procedures and findings are backed by flowcharts and summaries to ensure that understanding of the conducted optimization procedures is well enhanced.The results are useful to enhance deep learning models using higher optimization methods.