Authors:
Pavol Vančo
and
Igor Farkaš
Affiliation:
Comenius University, Slovak Republic
Keyword(s):
Recursive self-organizing networks, Tree structures, Output representation.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Methodologies and Methods
;
Neural Networks
;
Neurocomputing
;
Neurotechnology, Electronics and Informatics
;
Pattern Recognition
;
Physiological Computing Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
;
Theory and Methods
Abstract:
During the last decade, self-organizing neural maps have been extended to more general data structures, such as sequences or trees. To gain insight into how these models learn the tree data, we empirically compare three recursive versions of the self-organizing map – SOMSD, MSOM and RecSOM – using two data sets with the different levels of complexity: binary syntactic trees and ternary trees of linguistic propositions. We evaluate the models in terms of proposed measures focusing on unit’s receptive fields and on model’s capability to distinguish the trees either in terms of separate winners or distributed map output activation vectors. The models learn to topographically organize the data but differ in how they balance the effects of labels and the tree structure in representing the trees. None of the models could successfully distinguish all vertices by assigning them unique winners, and only RecSOM, being computationally the most expensive model regarding the context representatio
n, could unambiguously distinguish all trees in terms of distributed map output activation.
(More)