TY - JOUR
T1 - New algorithm for efficient pattern recall using a static threshold with the Steinbuch Lernmatrix
AU - Hernández, José Juan Carbajal
AU - Fernández, Luis P.Sánchez
PY - 2011/3
Y1 - 2011/3
N2 - An associative memory is a binary relationship between inputs and outputs, which is stored in an M matrix. The fundamental purpose of an associative memory is to recover correct output patterns from input patterns, which can be altered by additive, subtractive or combined noise. The Steinbuch Lernmatrix was the first associative memory developed in 1961, and is used as a pattern recognition classifier. However, a misclassification problem is presented when crossbar saturation occurs. A new algorithm that corrects the misclassification in the Lernmatrix is proposed in this work. The results of crossbar saturation with fundamental patterns demonstrate a better performance of pattern recalling using the new algorithm. Experiments with real data show a more efficient classifier when the algorithm is introduced in the original Lernmatrix. Therefore, the thresholded Lernmatrix memory emerges as a suitable and alternative classifier to be used in the developing pattern processing field.
AB - An associative memory is a binary relationship between inputs and outputs, which is stored in an M matrix. The fundamental purpose of an associative memory is to recover correct output patterns from input patterns, which can be altered by additive, subtractive or combined noise. The Steinbuch Lernmatrix was the first associative memory developed in 1961, and is used as a pattern recognition classifier. However, a misclassification problem is presented when crossbar saturation occurs. A new algorithm that corrects the misclassification in the Lernmatrix is proposed in this work. The results of crossbar saturation with fundamental patterns demonstrate a better performance of pattern recalling using the new algorithm. Experiments with real data show a more efficient classifier when the algorithm is introduced in the original Lernmatrix. Therefore, the thresholded Lernmatrix memory emerges as a suitable and alternative classifier to be used in the developing pattern processing field.
KW - Artificial intelligence
KW - Associative memories
KW - Classifier
KW - Neurocomputing
KW - Pattern processing
UR - http://www.scopus.com/inward/record.url?scp=79956212647&partnerID=8YFLogxK
U2 - 10.1080/09540091.2011.557716
DO - 10.1080/09540091.2011.557716
M3 - Artículo
SN - 0954-0091
VL - 23
SP - 33
EP - 44
JO - Connection Science
JF - Connection Science
IS - 1
ER -