Title :
TDNN with masked inputs
Author :
Alleau, Fabrice ; Poisson, Emilie ; Gaudin, Christian Viard ; Le Callet, Patrick
Author_Institution :
Sch. of Electron. & Comput. Eng., Polytech. Nantes, France
Abstract :
A novel architecture based on a time delay neural network is proposed in this paper. The main idea is the introduction of selection masks which allow to specify neurons to a subset of features included in their receptive field. It turns out that the number of free parameters can be decreased while the performances of recognition remains very high. This architecture has been designed for embedded online handwriting recognition systems where memory capacity has to be reduced as much as possible.
Keywords :
data encapsulation; handwriting recognition; neural nets; embedded online handwriting recognition system; masked input; memory capacity; neuron specification; time delay neural network; Artificial neural networks; Cellular neural networks; Character recognition; Computer architecture; Handwriting recognition; Hidden Markov models; Neural networks; Personal digital assistants; Switches; Testing;
Conference_Titel :
Information, Communications and Signal Processing, 2003 and Fourth Pacific Rim Conference on Multimedia. Proceedings of the 2003 Joint Conference of the Fourth International Conference on
Print_ISBN :
0-7803-8185-8
DOI :
10.1109/ICICS.2003.1292607
