3. Unsupervised Learning

Representation Learning — Quiz

Test your understanding of representation learning with 5 practice questions.

Practice Questions

Question 1

Which effect does adding a sparsity penalty to the hidden activations in a sparse autoencoder have on the learned representation?

Question 2

Which additional term does a contractive autoencoder include in its loss function to encourage robustness to small input perturbations?

Question 3

What is the primary rationale for using greedy layer-wise pretraining when training deep autoencoders?

Question 4

An autoencoder trained on normal data has a mean reconstruction error of $0.02$ with a standard deviation of $0.005$. Using a threshold of mean plus three standard deviations, what is the reconstruction error threshold for anomaly detection?

Question 5

Why might one apply K-means clustering to the latent representations learned by an autoencoder instead of the raw input data?