Automatic control learning systems (in the light of experiments on teaching the systems pattern recognition)

Summary The paper is devoted to the problem of teaching automatic systems pattern recognition. At the basis of the experiments that were conducted with regard to teaching, lies a profound and original hypothesis relating to the teaching process—a so-called ‘compactness hypothesis’. This hypothesis is interesting in that it opens a path to the mathematical description of the essence involved in the teaching process, which description is not attached to any concrete technical realization of the learning system nor to any concrete pattern type. At the same time, it prompted the experiments described in the present paper, and also made it possible to comprehend already-known studies on teaching. The fact that the experiments on teaching described in this paper were conducted on a universal (multi-purpose) computer is naturally tied in with the presence of such an hypothesis. The ‘compactness hypothesis’ makes it possible to formulate various teaching algorithms. Two of them (the algorithm of random planes and the algorithm of potential surfaces) were verified experimentally and yielded good results.