|
Article information
1999 , Volume 4, Special issue, p.55-68
Gorban A.N., Senashova M.Y.
Fast differentiation, duality and backward error propagation
Graphs are described for the calculation of composite functions of multiple variables, the duality principle for obtaining the gradient of a composite function is presented, a class of such functions is singled out for which the gradient can be most easily calculated (it comprises neuron networks but not only them, as there are extremely simple schemes of a more general form), fast algorithms for the evaluation of the calculation error are described.
Classificator Computer Science:- *G.1.4 Quadrature and Numerical Differentiation
- I.2.6 Learning
Author(s):
Gorban Aleksandr Nikolaevich
Office: Institut Computing Simulation of SB RAS
Address: 660036, Russia, Krasnoyarsk, Akademgorodok
Senashova M Yu
PhD.
Position: Senior Research Scientist
Office: Institut Computing Simulation of SB RAS
Address: 660036, Russia, Krasnoyarsk, Akademgorodok
E-mail: msen@icm.krasn.ru
Bibliography link:
Gorban A.N., Senashova M.Y. Fast differentiation, duality and backward error propagation // Computational technologies. 1999. V. 4. Special issue. P. 55-68
|
|
|
|
