Question from the Machine learning - Fundamentals test

It is used when developing a neuron network (because deep learning model doesn't learn the way machine learning one does)

It enables an agent to learn in an interactive environment by trial and error using feedback from its own actions

It combines supervised and unsupervised learning : it was invented to enable devs to be able to use both with on one single model

it is a paradigm and not a type of model: it can be made using a neural network, a decision tree, etc.