Machine Learning Concepts Quiz

What is the key mechanism that differentiates Transformers from RNNs?

• Convolutional layers

• Attention mechanism

• Sequential processing

• Recurrent connections

What are embeddings in the context of machine learning?

• Binary representations of data

• Compressed vector representations of data

• Hyperparameter optimization techniques

• Direct input-output mappings

Which of the following is true about Variational Autoencoders (VAEs)?

• They use adversarial loss to generate images

• They encode inputs into a latent space using a probabilistic distribution

• They are primarily used for classification tasks

• They do not include a decoder component

In Generative Adversarial Networks (GANs), the generator's goal is to:

• Classify real vs. fake data

• Minimize the discriminator's ability to distinguish real from fake data

• Increase the discriminator's loss

• Compress data into a latent space

What is a major limitation of GANs compared to VAEs?

• GANs are computationally inefficient

• GANs struggle with mode collapse

• GANs cannot generate high-quality images

• GANs are incapable of learning latent representations

Which neural network component is central to Stable Diffusion models?

• Transformer blocks

• Convolutional neural networks (CNNs)

• U-Net architecture

• Recurrent neural networks (RNNs)

What role does the 'noise' play in diffusion models like Stable Diffusion?

• It represents random perturbations added to training data to stabilize learning

• It represents the gradual transformation from random noise to coherent data

• It is used to regularize the model

• It acts as a latent embedding space

What type of optimization is used in Stable Diffusion to reconstruct data from latent noise?

• Gradient descent

• Variational inference

• Stochastic gradient descent (SGD)

• Denoising score matching

Which component in Transformers is responsible for capturing positional relationships in sequences?

• Feed-forward networks

• Attention weights

• Positional encodings

• Layer normalization

A real-life application of cross-attention in Stable Diffusion is:

• Generating images based on textual descriptions, such as 'a cat riding a bicycle.'

• Predicting stock prices from historical data.

• Identifying anomalies in network traffic.

• Clustering customers based on purchasing patterns.