HSM-0-0/smai-a4-mlp

MLP for CIFAR-10 (SMAI Assignment 4)

A 3072-256-10 network trained with plain SGD for 100 epochs on the 50,000 CIFAR-10 training images with true labels. Test accuracy: 0.5047. This is the plain MLP used as the baseline.

Files

• model.npz: weights W0, W1 and biases b0, b1, plus arch and acts. Each layer computes x @ W + b.
• preproc.npz: the training mean and std, one value per input column.
• config.json: architecture, activations, preprocessing and training settings.

Preprocessing

Flatten each image to 3072 values, divide by 255, subtract mean and divide by std.