Randomized learning in Y L-neural nets

The author proposes the perceptive artificial neural net, which consists of two different types of formal neurons with greatest simplicity. It is found that the net of the two simplest units (Y- and lambda -neurons) is capable of representing some classes of net topologies and can resolve the general categorization problem. It is also shown how to implement well-known statistical physics methods to learn the Y lambda -net, and some properties of learning trajectories are discussed. The simulated annealing technique is used at the learning session, and some advantages and disadvantages of this general method are pointed out.<
>