Quiz: Training Neural Networks: The Complete Learning Loop

Question 1 of 8

What is the first step in the complete training loop?

Initialize weights with random values

Forward propagation to make prediction

Calculate loss to measure error

Backward propagation to find responsible weights

Question 2 of 8

What does one epoch represent?

One complete pass through the entire training dataset

Processing one batch of images

One weight update

One forward propagation step

Question 3 of 8

In the example with 60,000 images and a batch size of 32, how many batches are there per epoch?

1,875

60,000

18,750

Question 4 of 8

What is the sweet spot range for batch sizes mentioned in the content? (Read through https://infolia.ai/archive/40 to answer)

32-128

1-10

256-512

1000-10000

Question 5 of 8

What technique is used when you stop training after validation loss stops improving?

Early stopping

Gradient clipping

Batch normalization

Dropout

Question 6 of 8

What are the classic metrics that indicate overfitting?

Training: 99%, Validation: 75%

Training: 60%, Validation: 58%

Training: 96%, Validation: 94%

Training: 50%, Validation: 50%

Question 7 of 8

What happens when the loss shoots up to NaN during training?

Learning rate is too high and the algorithm overshoots, hitting numerical instability

Learning rate is too low and training is too slow

The model is underfitting

The batch size is too small

Question 8 of 8

What is the recommended starting value for the learning rate hyperparameter?

0.001

0.1

0.0001

1.0