Mayank-Kar/smai-a4-mlp

SMAI Assignment 4 — MLP on CIFAR-10

Multi-layer perceptron baseline. Architecture 3072 → 32 → 10, hidden activation relu, 98,666 parameters, trained for 100 epochs on the true labels (SGD, lr 0.05, batch size 64). Final test accuracy 0.4216, train accuracy 0.5914.

Files: model.npz (weights for the NumPy library of the assignment, W stored as (n_in, n_out)), model.pt (PyTorch state_dict of an nn.Sequential), preprocessing.npz (per-column mean and std), config.json (architecture, training settings, preprocessing).

Preprocessing: flatten the 32×32×3 image to 3072 values (row, column, channel order), divide by 255, subtract mean and divide by std from preprocessing.npz.