Increase in Neural Network Training: Use of Mann Iteration for Deep Learning with Multi-Layer Perceptrons

This work examines how different machine learning models perform using the digits dataset, concentrating on Multi-Layer Perceptrons (MLPs) and Decision Trees. We contrast models with various hidden units and optimization methods, such as the Mann iteration theorem in conjunction with stochastic gradient descent (SGD). According to our findings, the accuracy of MLPs increases with the number of hidden units; the MLP model with 128 hidden units achieved an accuracy of 96.11%. This model’s accuracy was further enhanced to 96.94% using SGD and the Mann iteration theorem, proving the efficacy of sophisticated optimization methods. By contrast, the Decision Tree model’s accuracy of 84.16% was much lower. To get good performance in multi-class classification problems, the study emphasizes the significance of model complexity and optimization techniques. Comprehensive confusion matrices offer more information about the classification performance and possible areas for development. These results highlight the potential of MLPs optimized using cutting-edge approaches for precise digit classification and point to directions for further study, such as feature engineering, hyperparameter tweaking, and ensemble method investigation.