Shivansh521/smai-a4-mlp

SMAI Assignment 4: MLP on CIFAR-10

The MLP compared in Section 1.5 (trained on the true labels, p = 0).

• Architecture: 3072-32-10, hidden activation relu, 10 logits (identity), softmax cross-entropy loss
• Constraint: none
• Preprocessing: flatten the 32x32x3 image to 3072 values (X.reshape(-1, 3072)), divide by 255, then standardise every column with the mean and std of the 50,000 CIFAR-10 train images (preprocessing.npz)
• Test accuracy: 0.4131

weights.npz holds w_0, b_0, w_1, b_1, ... with w_i of shape (n_in, n_out); config.json has the full training settings.