Before any model can be built, the team must ___ data that represents the problem well.
To collect data means to gather the examples a model will learn from. Collect is the standard ML collocation, behind "data collection" pipelines. Gather up, pick up, and grab are informal and imprecise. Researchers discuss "collecting a representative dataset," so collect data is the precise, professional phrasing for acquiring the raw material on which any machine-learning model depends before training can begin.
2 / 5
Supervised learning requires people to ___ each example with the correct answer.
To label data means to attach the correct target value to each training example. Label is the precise term, behind "data labelling" and "labelled datasets." Name, title, and call do not carry the supervised-learning meaning. Teams hire "annotators to label data," so label the data is the correct collocation for providing the ground-truth answers that supervised models need in order to learn the mapping from inputs to outputs.
3 / 5
With labelled data ready, engineers will ___ the model on a GPU cluster.
To train a model means to optimise its parameters on data so it learns the task. Train is the foundational ML verb, behind "training loop" and "training data." Teach, school, and coach are human-learning metaphors not used technically. Researchers say "train the network for ten epochs," so train the model is the precise, essential collocation for the core process of fitting a machine-learning model to data.
4 / 5
After training, the team will ___ the model on a held-out test set.
To evaluate a model means to measure its performance using metrics on unseen data. Evaluate is the standard term, behind "model evaluation" and "evaluation metrics" like accuracy and F1. Judge, rate, and grade are vaguer and not the fixed ML phrase. Researchers "evaluate on the validation set," so evaluate the model is the precise collocation for assessing how well a trained model generalises to new examples.
5 / 5
Rather than train from scratch, they will ___ a pretrained model on their own data.
To fine-tune a model means to continue training a pretrained model on task-specific data to adapt it. Fine-tune is the precise ML term, central to transfer learning and LLM customisation. Adjust and tweak are too general, and tune up is about engines. Researchers "fine-tune on a downstream task," so fine-tune the model is the correct collocation for specialising a general pretrained model for a particular use case.