ML from zero: the one story
The picture: instead of programming rules by hand, you show a computer thousands of examples (emails labeled spam or not-spam) and it learns the pattern itself. The learned pattern is a model. The learning process is training. Using the model on new data is inference. Making it available to apps is deployment.
The four exam domains are literally the four life stages of a model: 1 prepare the data, 2 build the model, 3 deploy it, 4 monitor & secure it, then loop, because models go stale and get retrained. When a question confuses you, first ask which stage of the loop it lives in.
Three types of learning. Supervised: examples come with correct answers attached ("classify/detect/flag" is classification, predicting a category; "predict/estimate/forecast a value" is regression, predicting a number). Unsupervised: no answers, so you find structure, e.g. clustering customers. Reinforcement: trial-and-error with rewards, rare on the exam, just recognize it.
If the data contains the thing you're predicting, it's supervised. Category means classification; number means regression. The verb in the question tells you which.