Hints

DL-C06-C-001

Use prediction minus target.

DL-C06-C-002

Compute the error first, then square it.

DL-C06-C-003

Add the squared errors and divide by 3.

DL-C06-C-004

Class 3 means the third entry.

DL-C06-C-005

The maximum logit is 4.

DL-C06-C-006

One-of-many classification usually uses cross-entropy.