A Synergy of the Rosenblatt Perceptron and the Jordan Recurrence Principle

摘要

The recurrent perceptron RP-1 - a combination of the Rosenblatt perceptron and the Jordan network - is proposed. As a result, the perceptron acquires properties of recurrent neural networks, while the learning remains as fast and efficient as in the conventional Rosenblatt model. The recurrent perceptron is tested on the example of an agent education problem, where the "student" is placed into a model environment.