| |
Now that we have covered supervised machine learning, you should be able to define it with ease. If not, please take a step back and read the previous chapter. Apart from supervised machine learning, there are other concepts of machine learning like unsupervised machine learning that are gaining importance. In this technique, the machine is designed to interact with its ambient environment through actions. Based on the environment’s response to these actions, the machine receives rewards if the environment reacts positively or punishments if it reacts negatively. The machine will learn from the reactions, and this machine is taught to react in a manner by which it can maximize rewards and also identify futuristic events. The objective could also be to minimize future punishments. This technique of learning is related to the subjects of control theory in engineering and decision theory in statistics and management sciences.
The main problems studied in these two subjects are more or less equivalent, and the solutions are similar as well. However, both the subjects focus on different aspects of the problem. There is also another technique that uses unsupervised machine learning and game theory. The idea here is similar to that in unsupervised learning. The machine produces some actions that affect the surrounding environment, and it receives rewards or punishments depending on the reaction of the environment. However, the main difference is that the environment is not static. It is dynamic and can include other machines as well. These other machines are also capable of producing actions and receiving rewards (or punishments). So, the objective of the machine is to maximize its future rewards (or minimize its future punishments) taking into account the effects of the other machines in the surroundings.
The application of game theory to such a situation with multiple, dynamic systems is a popular area of research. Finally, the fourth technique is called unsupervised machine learning. In this technique, the machine receives training inputs, but it does not receive any target outputs or rewards and punishments for its actions. This begs the question - how can the machine possibly learn anything without receiving any feedback from the environment or having information about target outputs? However, the idea is to develop a structure in the machine to build representations of the input vectors in such a manner that they can be used for other applications such as prediction and decision-making. Essentially, unsupervised learning can be looked at as the machine identifying patterns in input data that would normally go unnoticed. Two of the most popular examples of unsupervised learning are dimensionality reduction and clustering. The technique of unsupervised learning is closely related to the fields of information theory and statistics.