Скачать или смотреть 3.2.Neural Network Foundations and Image Classification Exercises

This educational material focuses on a machine learning exercise involving image classification using the CIFAR-10 dataset. Students are guided through the implementation of a Multi-Layer Perceptron in Google Colab, where they explore how neural network architecture and data normalization impact performance. A significant portion of the assignment centers on weight visualization, transforming dense layer values into RGB images to see how the model perceives different classes. The sources further detail the effects of L2 regularization on preventing overfitting and how various activation functions like ReLU and sigmoid modify data. By comparing training and validation accuracy over multiple epochs, the exercise demonstrates practical methods for tuning and interpreting computer vision models.