Hierarchical Temporal Memory
Julia implementation of core Numenta HTM algorithms. Read the docs.
Hierarchical Temporal Memory is an abstract algorithmic model of the human brain (specifically the neocortex). It's a tool for
- neuroscience: understanding the human brain
- machine learning: predicting time series and detecting anomalies
The main algorithms of this model, the Spatial Pooler and Temporal (Sequence) Memory, are described in:
- Spatial Pooler
- Temporal Memory (section 3.3)
This package implements Numenta's Hierarchical Temporal Memory in simple and concise language, relying on linear algebra and staying close to the mathematical description in the source material.
Experiments and evaluation of this package lives in the HTMexperiments repo.
- Timeseries prediction tests, NAB results
- Explore temporal pooling ideas, influenced by forum discussions such as this.
- Maybe Backtracking TM? This non-biological spin on the TM algorithm hacks into the problem of learning repeating inputs. The only reference is the NUPIC implementation, focusing here. It isn't relevant to the current HTM research, only to applications.