Abstract:
In the paper, circuits obtained by superpositions over bases consisting of arbitrary subsets of particle-linear functions are studied, and criteria for the expressibility of an arbitrary particle-linear continuous function over such bases is proved. Estimates of the linear complexity and nonlinear dependence of the resulting circuits are given.
Key words:neural circuits, finitely parallel functions, expressibility of functions by neural circuits, nonlinear complexity, nonlinear depth.
Fulltext:
PDF file (241 kB)