WorksheetsAutoencoders and Generative Models Quiz
Total questions: 28
Worksheet time: 14mins
What is the primary purpose of an autoencoder?
Classification
Data compression and reconstruction
Supervised learning
Label prediction
Autoencoders are trained using which type of learning?
Supervised
Reinforcement
Unsupervised
Transfer
The compressed representation learned by an autoencoder is called:
Feature map
Hidden layer
Latent space
Output space
Which component of an autoencoder performs dimensionality reduction?
Decoder
Bottleneck
Encoder
Loss function
The bottleneck layer mainly helps to:
Increase accuracy
Prevent memorization
Increase model size
Speed up training
Latent variables are best described as:
Observable outputs
Noise variables
Underlying hidden variables
Labels
Which of the following is NOT a common application of autoencoders?
Image denoising
Anomaly detection
Facial recognition
Rule-based reasoning
Dimensionality reduction refers to:
Increasing features
Removing labels
Compressing data into fewer dimensions
Adding noise
A 28×28 grayscale MNIST image is represented as how many dimensions?
256
512
784
1024
Why is dimensionality reduction useful in ML?
Increases noise
Reduces accuracy
Improves efficiency and performance
Removes patterns
Which component reconstructs the original input?
Encoder
Bottleneck
Decoder
Optimizer
In a typical encoder, the number of neurons in each layer:
Increases
Remains constant
Decreases
Is random
The decoder architecture generally:
Mirrors the encoder
Has fewer layers
Uses no activation
Has constant size
VAEs differ from standard autoencoders because they are:
Deterministic
Supervised
Probabilistic
Rule-based
VAEs encode latent variables as:
Single values
Discrete labels
Probability distributions
Binary codes
In VAEs, latent variables are represented using:
Mean only
Variance only
Mean and standard deviation
Weights and bias
Reconstruction loss measures:
Model complexity
Distance between input and output
Training speed
Feature importance
KL divergence measures the difference between:
Two datasets
Two loss functions
Two probability distributions
Two neural networks
Generative AI primarily focuses on:
Classification
Prediction
Content creation
Rule execution
Which of the following is NOT a generative model?
VAE
GAN
Logistic Regression
Diffusion Model
Generative models learn:
Decision boundaries
Data distribution
Feature labels
Class margins
Discriminative models mainly learn:
Joint probability
Marginal probability
Conditional probability
Latent space
The core assumption of Naive Bayes is:
Features are dependent
Features are independent
Labels are continuous
Data is sequential
Maximum Likelihood Estimation (MLE) aims to:
Minimize error
Maximize accuracy
Maximize likelihood of observed data
Reduce parameters
Additive (Laplace) smoothing is used to:
Reduce noise
Increase variance
Avoid zero probabilities
Remove bias
Representation learning refers to:
Manual feature design
Learning rules
Automatically learning meaningful features
Data labeling
What is the main advantage of using a Variational Autoencoder (VAE) over a traditional autoencoder?
Faster training
Better reconstruction quality
Probabilistic interpretation of latent space
More complex architecture
In the context of neural networks, what does the term 'overfitting' refer to?
Model performs well on training data but poorly on unseen data
Model has too many parameters
Model is too simple to capture the underlying patterns
Model is unable to learn from the training data
