THERMODYNAMIC FORMALISM FOR THE STUDY AND FORMATION OF ALGORITHMS AND NEURAL NETWORKS
Free (open access)
Volume 1 (2013), Issue 4
456 - 465
I.F. YASINSKIY & F.N. YASINSKIY
This article considers the questions connected with creation of optimum algorithms using the laws of thermodynamics as applied to a computing process. Ideas and methods of phenomenological and statistic thermodynamics are used to estimate the amount of calculations or volume of the neural network. Introduction of the other thermodynamic functions, besides entropy, and also definition of the three thermodynamic origins in the context of calculations allow to study stability, organize the parameters according their information weights, carry out the decomposition of the complex systems, construct the rapid algorithms. The way of creation of the neural network structure is offered, consisting in use of the pre-trained fragments.
Computational entropy, neural network structure, rapid algorithms, thermodynamics of calculations