S. Štolc (Slovak Republic, Austria) and I. Bajla (Austria)
Machine learning, pattern recognition, hierarchical temporal memory (HTM), nonnegative matrix factorization (NMF)
In the paper we describe basic features of the computa tional intelligence network based on the Hierarchical Temporal Memory (HTM) that appeared recently in a form of research release of the system NuPIC (Numenta Platform for Intelligent Computing). The hierarchically structured HTM network makes use of spatio-temporal relations be tween training samples for generation of their invariant representations. There are several open issues for a research into HTM networks, in particular those applied to the visual pattern recognition tasks. In the paper, we report our results of the HTM architecture design and optimization of the network parameters for the task of recognition of the occluded gray-scale face images from the public ORL and Yale databases. We demonstrate that already very simple HTM architecture yields recognition rates that exceed the values achieved by recent vector subspace approaches based on nonnegative matrix factorization.
Important Links:
Go Back
