Hierarchical temporal memory

A hierarchical Temporalspeicher (Hierarchical Temporal Memory, HTM ) is a model of machine learning, which by Jeff Hawkins and Dileep George ( Numenta, Inc.) has been developed. This model makes several properties of the neocortex into a Bayesian network.

Design and function

HTMs are organized as a hierarchical network of nodes. Implements each node, as in a Bayesian network, a learning and memory function. The structure is constructed in such a way to create a hierarchical presentation of the data on the basis of time-varying data. However, this is only possible if the data in both the ( problem ) space, and in time are hierarchically represented directly.

A HTM performs the following functions, with the last two - depending on the implementation - are optional:

Criticism

HTMs provide for AI researchers is nothing fundamentally new, but are a combination of existing techniques, Jeff Hawkins but not sufficient points to the origins of his ideas. In addition, Jeff Hawkins, the usual in science publications for peer review and allow a sound check bypassed by scientists. It should however be noted that Hawkins does not come from academia, but from the industrial environment.