Reinforcement learning is part of the unsupervised learning space. Its goal is to make a model behave better and better, but we don't have the ground truth, a set of labeled data, for instance, to train our model. This only thing we can do is to use the network, and if the network gets a good result, then we use it to enhance our model with backpropagation. Otherwise, we try some more.
We can also use this approach in finance to optimize a portfolio; this can also be used for robots. In the past, people use genetic algorithms to train a walking robot, but now we can also use reinforcement learning for this task!
Now we have neural networks that can come to the rescue. Let's look at a few of the main types of networks that have been given attention in the last few years.