How does a machine learn?
Recently we hear a lot about artificial intelligence applications. However, probably you don’t know about the basics and how a machine can learn. Today I wanna talk about some fundamental machine learning methods.
Personally I like the toddler anology, when I talk about machine learning. Like a toddler a machine knows nothing at the beginning. But with time it gains experience and learns more about environment or about a specific topic. Let’s look at different learning methods.
1- Supervised Learning
In supervised learning both the input data and output data given. With enough training the system learns to match the input with the right output data. Assume that you wanna implement some system, which recognizes handwritten digits. Each people has a unique handwriting so each digit will be unique. We can prepare a data set, which consist of lots of handwritten digits, so our programm can train with this data and after it can predict the correct digit.
2- Unsupervised Learning
In this case the system uses only input data for training. It tries to categorize the data into different groups. As an example you can think of an onlineshop. You have lots of information about customers and sold products. Maybe you can implement a program which categorizes this data and optimizes your profit.
3- Reinforcement Learning
It seperates from both learning approaches above. In this case the machine learns from its own experiences with a rewarding system. You can think of a game. In some situations you should make the right move, otherwise you lose. From a perspective of machine it is a minus point, a bad experience in other words. So the machine learns from its own mistakes with each move.
This kind of learning process is used in robotics, automous driving and games.
I would be happy to see your constructive comments and questions.