Title :
Processing spatio-temporal patterns by mapping time into intensity
Author_Institution :
Mitsubishi Electr. Corp., Amagasaki, Japan
Abstract :
The author proposes a novel hybrid combination of networks for the processing of spatio-temporal patterns. While the instantaneous spatial components of patterns are translated into firing intensities of `grandmother´ cells, the temporal sequence of patterns is mapped into intensities of so-called TIM (time-intensity-mapping) cells. The TIM cells generate a corresponding (spatially distributed) activity pattern which in turn is processed by another set of grandmother cells. The feasibility of the approach is demonstrated by using temporal sequences of letters. The robustness against distortions of transient pattern sequences is remarkable and may lead to applications of the TIM network in speech recognition
Keywords :
neural nets; pattern recognition; firing intensities; grandmother cells; instantaneous spatial components; neural nets; robustness; spatio-temporal patterns processing; speech recognition; temporal sequences; time-intensity-mapping; transient pattern sequences; Information processing; Laboratories; Matched filters; Neural networks; Pattern matching; Pattern recognition; Sampling methods; Spatiotemporal phenomena; Speech processing;
Conference_Titel :
Neural Networks, 1991., IJCNN-91-Seattle International Joint Conference on
Conference_Location :
Seattle, WA
Print_ISBN :
0-7803-0164-1
DOI :
10.1109/IJCNN.1991.155449