Autoencoders and Generative Models

Five questions. Choose one answer for each, then submit to see your score and explanations.

The practice target is 80%. You can retry as often as you like, and every lesson stays accessible.

You can practice as a guest. Sign in before starting an attempt to save your scores.

1. What is a basic autoencoder trained to do?A. Guarantee useful random samples from any latent pointB. Reconstruct its input through an encoded representationC. Predict every future label with no assumptionsD. Assign a unique class name to each pixel

2. Why can an autoencoder bottleneck be useful?A. It constrains the representation and encourages useful structureB. It guarantees the smallest file for any inputC. It removes every possible reconstruction errorD. It makes all training data unnecessary

3. What is a key objective of a generative model?A. Only assign labels to a fixed set of imagesB. Always reproduce the training set pixel for pixelC. Prove that any output is factually correctD. Learn data structure well enough to produce plausible new samples

4. An autoencoder trained on normal parts reconstructs one damaged part poorly. What is a justified interpretation?A. Any bright image must be defectiveB. The model no longer needs representative normal examplesC. The error can be an anomaly signal that needs a validated thresholdD. The part is definitely damaged with 100% certainty

5. A generator makes attractive images but only one kind of example. What should evaluation check?A. Whether the model name contains a popular keywordB. Diversity as well as sample qualityC. Only whether one output looks realisticD. Only the size of the download file

Review this module from its first lesson →

Autoencoders and Generative Models

Your result