Hierarchical Temporal Memory for Behavior Prediction

Abstract

This thesis is about researching and analyzing Hierarchical Temporal Memory, specifically the newly developed "HTM Cortical learning algorithms"[3] developed by Jeff Hawkins and the company Numenta. Two problems are addressed. Can this type of hierarchical memory system make an internal representation of simple data sequences at the input? And if so, does it take long to learn? Two C++ applications were developed in this thesis. The first program is used to analyze the algorithm, and the second program is used to visualize the internal states of the network. The results is very dependent of how the system is configured. If enough resources are available, the system can learn sequences, and it does not take long for the system to learn.