Awesome

ARTimeNAB

This is an open source (AGPL3) and simplified version of ARTime, an anomaly detection algorithm. It supports the Numenta anomaly benchmark (NAB) ARTime detector.

NAB includes the ARTime detector, please start there to see & reproduce the ARTime results. The NAB environment uses PythonCall to install ARTime from this repository.

ARTime was developed by Mark Hampton.

Running ARTime with NAB

The Python JuliaCall module is used with NAB and the version of JuliaCall we are using defaults to installing the latest stable version of Julia (ignoring the Julia version in the juliacalldeps.json file at the root of NAB). ARTime is no longer compatible with the most recent Julia language. To use Julia 1.7.0 with JuliaCall you must install Julia 1.7.0 and set the environment variable: PYTHON_JULIACALL_EXE to the Julia 1.7.0 binary executable before running the ARTime detector.

There is a fork of the NAB repo with a docker environment that runs ARTime at https://github.com/markNZed/NAB/tree/docker with a README

Acknowledgements

Stephen Grossberg and Gail Carpenter developed adaptive resonance theory (ART). Grossberg's 2021 book Conscious Mind, Resonant Brain: How Each Brain Makes a Mind was the major inspiration for ARTime.

Numenta provided NAB to inspire innovation in anomaly detection. It was very valuable in testing ARTime. The paper introducing NAB is from Ahmad, S., Lavin, A., Purdy, S., & Agha, Z. (2017). Unsupervised real-time anomaly detection for streaming data. Neurocomputing, Available online 2 June 2017, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2017.04.070

The excellent Julia package AdaptiveResonance.jl was extremely useful in getting ARTime off the ground. Modifications in the DVFA implementation of AdaptiveResonance.jl led to a compact version of AdaptiveResonance being included in ARTime.

@isentropic was a great help in introducing me to Julia and improving the quality of the code.

Where to from here

Unfortunately deep learning catastrophically forgot why it was not a panacea in the 1980s. A talented team of computational neuroscientists could push ART much further in machine learning... For an introduction to Grossbergian Neuroscience look no further than Yohan John's Neurologos channel on YouTube.