Refining Computational Models with the Development of Efficient Big Data Models, Algorithms, and Architectures

Big Data has prompted a new approach to improving computer models: the Enhanced Parallelized Neural Network (EPNN). The three primary components of EPNN are P-NN, AFS, and HCFEA. Parallelization helps PNN overcome the computing power shortage that plagues neural network training. This strategy reduces massive dataset training time by changing model parameters on numerous machines at once. Fig. 1 shows P-NN's steps, demonstrating its ease of handling vast volumes of data. AFS recommends dynamic feature selection for model training. Figure 2 demonstrates how partial derivatives of the loss function dynamically modify feature weights. This makes the model more usable, understandable, and less impacted by unimportant aspects. To maximize computing performance, HCFEA uses a hybrid cloud-fog edge design. The task assignment mechanism in Figure 3 modifies jobs depending on data proximity to decrease latency. EPNN objective function sections address neural network loss, adaptive feature selection, and delay minimization. EPNN consistently outperforms well-known approaches in several factors. Table 2 reveals that EPNN trains quicker, is more accurate, and is more scalable than other approaches. EPNN's flexible feature selection, dynamic design efficiency, and decreased latency make it a preferable choice (Table 3). The pie chart, scatter plot, histogram, and heat map in Figures 4, 5, 6, and 7 indicate EPNN's superiority in training time distribution, overall performance parameters, score frequency distribution, and whole performance comparison.