Question$-1(50$ points$)$:

Take our CIFAR classification example:

a$)(10$ points$)$ Measure train speed when running in CPU and GPU. HINT: Measure how long

it takes to train one iteration.

b$)(40$ points$)$ Change the classification head as follows:

Use a linear transformation that transforms features to $200$

Use Tanh

Use a linear transformation that transforms features to $100$

Use Relu

Use a linear transformation that transforms features to $50$

Use Softmax

Run training for $5$ epochs with two different learning rates. Measure the accuracy for each.