Virtual Labs

Autoencoders for Representation Learning

What is the primary objective of an autoencoder?

a: To classify images into different categories Explanation

Explanation

b: To learn a compressed representation by reconstructing input data Explanation

Explanation

c: To predict future values in a time series Explanation

Explanation

d: To generate new images from random noise Explanation

Explanation

Which component of an autoencoder compresses the input into a lower-dimensional representation?

a: Decoder Explanation

Explanation

b: Classifier Explanation

Explanation

c: Encoder Explanation

Explanation

d: Pooling layer Explanation

Explanation

What is the bottleneck layer in an autoencoder?

a: The layer with the smallest dimension that holds the compressed representation Explanation

Explanation

b: The input layer with maximum dimensions Explanation

Explanation

c: The final output layer Explanation

Explanation

d: A layer that prevents overfitting Explanation

Explanation

In a denoising autoencoder, what is used as input during training?

a: Clean images only Explanation

Explanation

b: Noisy or corrupted images Explanation

Explanation

c: Random noise vectors Explanation

Explanation

d: Latent representations Explanation

Explanation

What type of learning paradigm do autoencoders belong to?

a: Supervised learning Explanation

Explanation

b: Reinforcement learning Explanation

Explanation

c: Unsupervised learning Explanation

Explanation

d: Semi-supervised learning Explanation

Explanation

What loss function is typically used to train autoencoders?

a: Cross-entropy loss Explanation

Explanation

b: Hinge loss Explanation

Explanation

c: Kullback-Leibler divergence only Explanation

Explanation

d: Mean squared error or reconstruction loss Explanation

Explanation

What is the size of each image in the Fashion-MNIST dataset?

a: 32×32 pixels Explanation

Explanation

b: 28×28 pixels Explanation

Explanation

c: 64×64 pixels Explanation

Explanation

d: 224×224 pixels Explanation

Explanation

What advantage does a 2-D latent space provide in autoencoders?

a: Faster training speed Explanation

Explanation

b: Better reconstruction quality always Explanation

Explanation

c: Easy visualisation of learned representations Explanation

Explanation

d: Elimination of all reconstruction errors Explanation

Explanation

What happens if the bottleneck dimension is too small in an autoencoder?

a: The model may fail to capture important features, leading to poor reconstruction Explanation

Explanation

b: Training becomes faster Explanation

Explanation

c: The model will always overfit Explanation

Explanation

d: Latent space becomes easier to interpret Explanation

Explanation

What is the main advantage of using autoencoders over supervised learning methods for feature extraction?

a: Autoencoders always achieve 100% accuracy Explanation

Explanation

b: Autoencoders do not require labelled data and can learn representations from raw data alone Explanation

Explanation

c: Autoencoders can only work with text data Explanation

Explanation

d: Autoencoders eliminate the need for training Explanation

Explanation