Flashcards on Generative Models: LLMs and Transformers

Generative Models: LLMs & Transformers Explained for Students

1 / 20

What are the two main learning paradigms in machine learning mentioned in the content?

Supervised learning and unsupervised learning.

Tap to flip · Swipe to navigate

Machine Learning Paradigms

20 cards

Card 1

Question: What are the two main learning paradigms in machine learning mentioned in the content?

Answer: Supervised learning and unsupervised learning.

Card 2

Question: Give three examples of supervised learning algorithms listed.

Answer: Linear regression, support vector machines (SVM), and logistic regression.

Card 3

Question: Give three examples of unsupervised learning methods listed.

Answer: k-Means, Gaussian mixtures, and Principal Component Analysis (PCA).

Card 4

Question: What is the general objective in supervised learning as described?

Answer: Minimize a loss function combined with a regularization/penalty term (loss + regularization).

Card 5

Question: Which loss corresponds to a Gaussian likelihood and which to a Bernoulli likelihood?

Answer: Squared loss corresponds to a Gaussian likelihood; logistic loss corresponds to a Bernoulli likelihood.

Card 6

Question: What regularizers correspond to Gaussian and Laplace priors?

Answer: L2 norm corresponds to a Gaussian prior; L1 norm corresponds to a Laplace prior.

Card 7

Question: Name two robust or alternative loss types mentioned besides squared and logistic loss.

Answer: 0/1 loss and hinge loss (also student-t likelihood and cost-sensitive losses are mentioned).

Card 8

Question: List three optimization or training methods referenced.

Answer: Gradient Descent (including stochastic/minibatch SGD), EM algorithm, and Lloyd’s heuristic for k-Means.

Card 9

Question: What are common model selection techniques named in the content?

Answer: K-fold cross-validation, Monte Carlo cross-validation, and Bayesian model selection.

Card 10

Question: What methods are listed for deriving models or solutions?

Answer: Exact solutions (e.g., eigendecomposition for PCA), gradient descent/SGD, reductions, EM, Lloyd’s heuristic, and Bayesian model averaging.