Hyperparameter Tuning and Neural Architecture Search
Total points 3
1.
Question 1
Neural Architecture Search (NAS) was a promising technique that failed to surpass hand-designed architectures in terms of test set accuracy.

1 / 1 point

True


> False

Correct
Spot on! In fact, NAS can design a novel network architecture that rivals the best human-invented architecture.

2.
Question 2
Which of the following characteristics best describe hyperparameters? (Select all that apply)

1 / 1 point

> Hyperparameters can be quite numerous even in small models.

Correct
Great job! Hyperparameters can be numerous, so, performing manual hyperparameter tuning can be a real brain teaser.


> Hyperparameters are set before launching the learning process.

Correct
Excellent! They need to be set before model training begins.

> Hyperparameters are not optimized in each training step.

Correct
You’re right on track! Hyperparameters are not automatically optimized during the training process.

Hyperparameters are derived via training.

You didn’t select all the correct answers
3.
Question 3
Does KerasTuner support multiple strategies?

1 / 1 point

> Yes


No

Correct
Exactly! KerasTuner comes with Bayesian Optimization, Hyperband, and Random Search algorithms built-in.