Supervised Learning
Learning from examples that come with the correct answer, so the model can predict answers for new, unlabeled examples.
Alsoclassification · regression
Learning with flashcards. The front shows a picture, the back says 'dog' or 'cat'. After enough cards you can name animals on cards you have never seen.
Where it fits
Not shown: 7 more that build on it. See the full list below.
+All relationships as a list
Is a kind of / part of
Siblings
Techniques that apply here
How it works
In supervised learning every training example is a pair: an input (the features) and the label (the right answer). The model learns a function from inputs to labels, then applies it to new inputs.
Two flavors:
- Regression: the label is a number. Price, temperature, time to delivery. Classic model: linear regression.
- Classification: the label is a category. Spam or not, which digit, which animal. Classic models: logistic regression, decision trees.
Training follows the foundations recipe: measure the mismatch between predictions and labels with a loss function, then adjust the model to reduce it. The real test is performance on examples held back from training.
Everyday use cases
- Case 01
Spam filtering
Emails labeled spam or not spam teach a classifier to sort new mail.
- Case 02
Price prediction
Past sales with their prices teach a regression model to value new listings.
- Case 03
Medical diagnosis support
Scans labeled by doctors teach a model to flag suspicious regions.
Strengths, weaknesses and when to use it
Strengths
- Clear goal and clear evaluation, because you know the right answers.
- The most mature paradigm, with the widest choice of models.
Weaknesses
- Labels are often expensive, slow or require experts.
- The model can only be as good, and as unbiased, as its labels.
When to use itWhen you have (or can afford to create) examples with the correct answer for the thing you want to predict.
Compare with
Check yourself
- Q01
Predicting tomorrow's temperature in degrees is which kind of supervised task?
Show answer
B: Regression
- Q02
What do you need for supervised learning that you do not need for clustering?
Show answer
B: Labels