Chapter C19 · Interactive Course
Different jobs for different models
Describe a model by the job it does, not by its design.
Core Learning Concept
Classify each service by the output its task needs: a category, quantity, order, grouping, unusual-case signal, or new content. A product can combine several of these jobs. Use the supplied descriptions to explain the task before considering a model name or a marketing label.
What You’ll Learn in This Chapter
-
Learning from data is one part of AI
Place machine learning inside the broader AI field.
Machine learning is an approach in which models learn patterns from data. AI is the broader field and also includes approaches based on explicitly designed rules or search. Identifying a learned component tells you to ask about its examples and testing, not just the instructions someone wrote.
-
Choosing a category
Recognize classification from its output.
Classification assigns an input to one or more defined categories. A parcel system might label an item “fragile” or “not fragile.” Some classifiers also return scores, but a score is not automatically a reliable probability. Check what the categories and scores mean before using the result.
-
Estimating a number
Recognize regression from its output.
Regression estimates a numerical quantity, such as delivery duration. An error has a size: predicting twenty minutes for an eighteen-minute trip misses by two minutes. Ask about units and typical error, because a precise-looking estimate may still be too uncertain for the decision you want to make.
-
Putting options in order
Recognize ranking as a model task.
Ranking orders candidates according to a chosen objective. A depot list could prioritize shortest journey, lowest price, or earliest availability, producing different orders. The ranking tells you what comes first under that objective. It does not establish that the first option is best for every purpose.
-
Finding groups
Recognize clustering without supplied category labels.
Clustering groups examples by similarity without requiring category labels. People inspect the discovered clusters to understand their patterns. Because similarity depends on the method and features chosen, different setups produce different groupings.
-
Noticing something unusual
Distinguish anomaly detection from proof of a problem.
Anomaly detection identifies examples that differ from an expected pattern. An unusually heavy parcel may deserve inspection without being faulty. The output is a signal to investigate, not proof of a problem. Check both unusual harmless cases and important problems that resemble ordinary cases.
-
Making new content
Recognize a generative task by the shape of its output, across media.
Generative models create content such as text, images, or audio. Judge generated outputs against task goals and verify factual statements carefully, because composing fluent language is distinct from establishing factual truth.
-
One model, several descriptions
Separate task, architecture, learning method and medium.
Different technical terms describe distinct aspects of a system. Neural network describes internal architecture, supervised names the training setup, and classification identifies the user-facing task. Keep these separate dimensions distinct.
-
Label a fictional AI service
Identify a service's output task, distinguish that from its design, and explain one limitation of the result.