Reinforcement training

Reinforcement learning (eng. Reinforcement learning ) is one of the ways of machine learning, during which the subject system ( agent ) learns by interacting with a certain environment . From the point of view of cybernetics, is one of the types of cybernetic experiment. The response of the environment (and not the special reinforcement management system, as it happens in teacher training) to the decisions made is reinforcement signals , so such training is a special case of teacher training, but the teacher is the environment or its model. You also need to keep in mind that some reinforcement rules are based on implicit teachers, for example, in the case of an artificial neural environment, on the simultaneous activity of formal neurons, because of which they can be attributed to learning without a teacher.

Medium and agent

The agent acts on the medium, and the medium acts on the agent. Such a system is said to have feedback. Such a system should be considered as a whole, and therefore the line of separation between the medium and the agent is rather arbitrary. Of course, from the anatomical or physical points of view there is a definite border between the medium and the agent (organism), but if this system is viewed from a functional point of view, the separation becomes fuzzy. For example, the cutter in the sculptor’s hand can be considered either as part of a complex biophysical mechanism that gives shape to a piece of marble, or as part of the material that the nervous system is trying to control.

Reinforcement system and its types

Rosenblatt tried to classify the various learning algorithms, calling them reinforcement systems. [1] He gives the following definition:

A reinforcement system is any set of rules, based on which you can change over time the interaction matrix (or memory state) of a perceptron.

In addition to the classical perceptron teaching method, the error correction method, which can be attributed to teaching with a teacher, Rosenblatt also introduced the concept of teaching without a teacher, suggesting several ways of teaching:

• The alpha reinforcement system is called the reinforcement system, in which the weights of all active bonds that lead to the element , change by the same value of r, and the weights of inactive bonds do not change during this time.

• The gamma system of reinforcement refers to such a rule for changing the weights of a certain element, in which the weights of all active bonds are first changed by an equal amount, and then another value is subtracted from their all bond weights equal to the total change in weights of all active bonds divided by the number of all bonds. This system has the property of conservatism with respect to the weights, since it has the full sum of the weights of all the connections that can neither increase nor decrease.