Title :
A hardware training device for forward feed hidden neural networks
Author :
Crilly, Paul Benjamin ; Pethel, Shawn D. ; Bowden, Charles M.
Author_Institution :
Dept. of Electr. & Comput. Eng., Tennessee Univ., Knoxville, TN, USA
Abstract :
A hardware device for training multi-hidden-layer neural networks is presented. It is based on a recently developed training algorithm that is an alternative to backpropagation. The new algorithm employs pseudoinverse techniques and preconditioning of the weights. The hardware device is implemented as an accelerator board for a personal computer or workstation and does the calculations via a ROM lookup table
Keywords :
feedforward neural nets; learning (artificial intelligence); neural net architecture; ROM lookup table; accelerator board; forward feed hidden neural networks; hardware training; personal computer; preconditioning; pseudoinverse matrix; pseudoinverse techniques; training algorithm; workstation; Backpropagation algorithms; Equations; Feedforward neural networks; Feeds; Instruments; Missiles; Neural network hardware; Neural networks; Neurons; Weapons;
Conference_Titel :
Instrumentation and Measurement Technology Conference, 1993. IMTC/93. Conference Record., IEEE
Conference_Location :
Irvine, CA
Print_ISBN :
0-7803-1229-5
DOI :
10.1109/IMTC.1993.382569
