Solutions

DL-C05-C-001

(16, 5) x (5, 9) -> (16, 9)

The bias broadcasts across the batch, and ReLU keeps the shape. So H has shape (16, 9).

DL-C05-C-002

(16, 9) x (9, 4) -> (16, 4)

The score matrix has shape (16, 4).

DL-C05-C-003

The network has one hidden layer and one output layer. Both have learned parameters, so it has 2 learned layers.

DL-C05-C-004

ReLU maps:

[2, -5, 1] -> [2, 0, 1]

There are 2 positive values.

DL-C05-C-005

The model produces one number per example. For 25 examples, the output shape is (25, 1).

DL-C05-C-006

The classifier produces 6 scores per example. For 25 examples, the score matrix has shape (25, 6).