Neural Networks for MNIST Digit Recognition The classic MNIST dataset consists of 28x28 pixel (784 features) grayscale images of handwritten digits 0-9 (10 classes).
In this problem, I implemented a neural network to classify the MNIST dataset using raw pixels as features. Specifically, I implemented a type of architecture called a Multi-Layered Perceptron (MLP), which consists of several Dense layers (i.e. “Perceptrons”) connected sequentially, with nonlinear activation functions after each layer. Then I trained this network using the Categorical Cross-Entropy loss function, usign SGD and Adam optimizer