How a machine learns
Level wrap-up: all of supervised learning fits into data + model + loss + optimizer
In this chapter, you will learn to
- See that every supervised model comes down to four ingredients: data, model, error to minimise, optimisation method
3 exercises
Each exercise is graded automatically, in your browser. The full statements are available once you have an account.
What this is for in real life
Behind products that have nothing in common, the same recipe. Once you see it, every new model becomes easier to understand. - Assistants like ChatGPT. Their first training phase consists in guessing the next…
The four ingredients
You have built several predictors: a line to estimate a price, a classifier to say "admitted / rejected"… On the surface, different problems. In reality, they all follow the same blueprint. Supervised…
The odd one out
We have just seen that most supervised methods = model + loss + optimizer: we tune parameters to minimize an error. But one of the methods you have encountered trains no parameters at all: it minimizes no loss…