MNIST MLP Inference Visualization is an interactive tool for observing the inference process of a handwritten digit recognition model in 3D. When you draw a digit on a 28 by 28 grid, you can see how activations propagate through each layer of a trained multilayer perceptron (MLP) and become final prediction probabilities.
Neuron activations, strong weighted connections, prediction distributions, and the training timeline are shown in the same view, making the internal representation of a neural network easier to understand visually. It is useful for learning image classification inference beyond just the input and output.
