Machine Learning Vocabulary Quiz
12 multiple-choice questions on machine learning vocabulary: supervised learning, training data, hyperparameters, precision and recall. B2–C1 level.
This quiz focuses on how the target vocabulary for Machine Learning is actually used in context at B2–C1 level, rather than testing bare definitions. Correct answers you will need to identify include terms such as supervised, training data, overfitting, hyperparameter and precision, each embedded in a full example sentence so you have to judge meaning from context, not just recognise an isolated word.
Working through all 12 questions and checking the explanations in the FAQ below is a quick way to spot any terms you are still unsure of. Revisiting the quiz again after a day or two, rather than only once, is one of the most reliable ways to move new vocabulary from passive recognition into words you can use confidently yourself in speaking and writing.
Keep building your machine learning vocabulary.
Machine Learning Vocabulary — FAQ
What is the difference between 'supervised' and 'unsupervised' learning?
Supervised learning is a type of machine learning where the model learns from labelled training examples, meaning each input has a known correct answer. Unsupervised learning, by contrast, finds patterns in data without any labels, often used to discover hidden clusters or groupings that were not defined in advance.
What is 'training data'?
Training data is the dataset used to teach a machine learning model how to make predictions. The quality and quantity of training data has a huge impact on how accurate and reliable the resulting model will be.
What does 'overfitting' mean?
Overfitting occurs when a model learns the training data too closely, including its noise, so it performs very well on that data but poorly on new, unseen examples. Techniques like regularisation and cross-validation are used to detect and reduce overfitting.
What is a 'hyperparameter'?
A hyperparameter is a configuration setting that is chosen before training begins and controls how the model learns, such as the learning rate or the number of layers in a neural network. Unlike the model's internal weights, hyperparameters are not learned automatically from the data.
What is the difference between 'precision' and 'recall'?
Precision is the proportion of positive predictions made by a model that are actually correct, so high precision means few false positives. Recall is the proportion of actual positive cases that the model correctly identifies, so high recall means few missed cases; in medical diagnosis, high recall is especially critical.
What is 'backpropagation'?
Backpropagation is the algorithm used to calculate how to update the weights of a neural network during training, by computing the gradient of the loss with respect to each weight. It is the foundation of how most modern neural networks learn from their mistakes.
What is an 'embedding'?
An embedding is a numerical representation of a word, image or other object as a vector, allowing a model to work with it mathematically. Word embeddings, for example, allow a model to capture semantic relationships between terms, such as recognising that 'king' and 'queen' are related concepts.
What does 'cross-validation' mean?
Cross-validation is a technique for evaluating a model by testing it on multiple different subsets of the available data, rather than a single train-test split. This helps ensure the model performs consistently well on unseen data rather than just on one lucky sample.
What is the difference between 'training' and 'inference'?
Training is the process of fitting a machine learning model to a dataset, adjusting its internal parameters so it can make accurate predictions. Inference is the later process of using that already-trained model to make predictions on new data, which typically needs to run much faster than training.
What is a 'loss function'?
A loss function is a measure of how far a model's predictions are from the correct answers during training. A lower loss function value means the model is making fewer errors, and training algorithms like gradient descent work by trying to minimise this value.