MatlabHTM: A sequence memory model of neocortical layers for anomaly detection

摘要

Many models based on the operation of the neocortex, which is the center of brain intelligence, are emerging. The Hierarchical Temporal Memory (HTM) model is a unique intermediate level model of the neocortex’s layered substructures. The hypothesis is that these layers build temporal models of sequences of observations and/or motor signals, i.e., build a sequence memory. Implementations of this model exist in Python, C++ and Java. However, those implementations are quite cumbersome to use, as they depend on many other packages. This paper presents a lean, standalone, easy to modify MATLAB implementation. The performance results from processing the Numenta Anomaly Benchmark (NAB) demonstrate the fidelity of matlabHTM.