Machine learning (ML) develops statistical algorithms that learn from data and generalize to unseen data to perform tasks without explicit programming. This is the definition of ML.

Context: Tests recall of the core ML definition concept name.

ML is grounded in and mathematical optimization methods, often framed as minimizing a loss function.

Context: Tests understanding of ML foundations.

A measures discrepancy between model predictions and actual outcomes.

Context: Tests the meaning of loss function.

Many ML/deep learning algorithms can be described as minimizing empirical risk under a probably approximately correct learning framework. This viewpoint is called (ERM).

Context: Tests ERM concept name and meaning.

is the ability to perform accurately on new, unseen examples after training on a dataset drawn from an unknown probability distribution.

Context: Tests the meaning of generalization.

Finite training sets and uncertain future which leads to learning theory providing probabilistic (not absolute) performance guarantees.

Context: Tests a cause→effect relationship word/phrase.

In the PAC learning framework, provides probabilistic bounds on performance given finite training data.

Context: Tests PAC learning concept name.

Supervised learning trains models using to predict outputs, commonly for classification or regression.

Context: Tests the meaning of supervised learning setup.

Unsupervised learning discovers structure in data, such as grouping via clustering or reducing dimensionality.

Context: Tests the meaning of unsupervised learning setup.

Advances in deep learning cause which leads to improved performance on many tasks.

Context: Tests a cause→effect relationship from the knowledge base.

Neural networks are statistical algorithms inspired by neuron interactions, and deep learning achieves strong performance in many tasks. This idea is part of .

Context: Tests the core concept name for neural networks and connectionism.

An algorithm for training neural networks by propagating error gradients is called .

Context: Tests the meaning of backpropagation.

A system predicts posterior probabilities of a sequence given its history. This which leads to optimal data compression.

Context: Tests a cause→effect relationship word/phrase.

Predicting posterior probabilities of sequences enables optimal data compression, and conversely an optimal compressor can be used for prediction. This relationship is called .

Context: Tests the meaning of compression–prediction equivalence.

Data mining (KDD) focuses on discovering previously unknown properties, while ML focuses on prediction of known knowledge learned from training data. This contrast is between and machine learning.

Context: Tests understanding of the data mining vs ML relationship.