Visualizing the Training Process of MLP

Abstract

Multilayer Perceptrons (MLPs) are extensively used in many machine learning applications; yet, because of their high dimensionality and complexity, their training process is still largely opaque. In order to better understand how MLPs learn and modify their internal representations, this study focuses on visualizing the evolution of decision boundaries during training. We gain insights into neural networks' learning dynamics and convergence behavior by analyzing how the decision regions change as training progresses. This work also contributes to the broader field of Explainable AI (XAI) by offering a visual approach to interpreting neural network decision-making, addressing the common critique of MLPs being black-box models.