A Hierarchical Temporal Memory Model in the Sense of Hawkins

Jeff Hawkins introduced a revolutionary Hierarchical Temporal Memory (HTM) concept in his 2004 book “On Intelligence”. He stated that previous attempts to create artificial intelligence are doomed to failure since they do not take into account biophysical processes in the brain. Although now we see some progress in deep neural networks for pattern recognition tasks, special bio-inspired models can be applied in the analysis of temporal signals and the search for anomalies. The proposed approach is based on the concept of memory-prediction when for each incoming signal, a hypothesis about its place in the general structure of signals is constructed. In this work, we review the method and evaluate its applicability, as well as implement a visualizer for HTM networks and then use this method to build a graph-based memory form from input MIDI sequences. Our implementation allows to create simple digital representations of musical compositions and can be used to search for similar melodies.