DSA Studio

Neural Network Abstractions Checkpoint

Test your understanding of Module, Linear, and Sequential.

Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. Linear(4,3,bias=True) is used twice with shared weights. How many distinct trainable scalars belong to that layer?

Select one answer.

2. A shared parameter is found through two attribute paths. How should an optimizer enumerate it?

Select one answer.

3. A layer uses uniform bounds +/-1/sqrt(fan_in). Which description is accurate?

Select one answer.

4. A bias-free Linear(2,5) has how many scalar parameters?
5. Calling eval() on these compact modules changes the mode flag. Does it automatically stop gradient tracking?

Select one answer.

Join Discord