I am pleased to announce that we published two new papers this month. The
first is an article that our co-founder Jeff Hawkins was invited to write for an
IEEE Spectrum Special Report titled “Can we Copy the Brain?”
The report, which focuses on worldwide efforts to understand the brain
and use that knowledge to enable the next generation of computing,
contains a number of articles by industry experts.
Jeff’s article, “What Intelligent Machines Need to Learn from the Neocortex,”
covers the importance of building biologically based intelligent
machines, three features of intelligence we can’t ignore, and the far
reaching implications of machine intelligence. This piece was
particularly exciting for us because it is the first time we’ve
published details on our latest discovery in sensorimotor integration.
Our research team is currently working on a peer-reviewed paper on
this topic, so expect to hear more about that in the coming months. In
the meantime, we hope even non-technical readers will enjoy Jeff’s IEEE
piece. I also sat down with Jeff to talk with him about this article.
You can listen to our conversation here.
The second paper published this month is a peer-reviewed paper titled
“Unsupervised real-time anomaly detection for streaming data”
that appears in a special issue of the journal Neurocomputing.
This paper, authored by Subutai Ahmad, Alexander Lavin, Scott Purdy and
Zuha Agha, highlights the importance of anomaly detection for streaming
applications and introduces two contributions from Numenta within that
field. First, it demonstrates how our online sequence memory algorithm,
HTM, excels at finding anomalies in streaming, time-series data because
it displays the characteristics of an ideal detector: it can process
data in real time, learn continuously and is fully automated. Second, it
presents the results of running selected algorithms, including HTM, on
the Numenta Anomaly Benchmark(NAB).
First released in late 2015, NAB is an open source benchmark that contains
real-world data streams with known, labeled anomalies. It was created to
provide a controlled yet open source tool that allows anyone to test anomaly
detection algorithms on streaming data.
Anomaly detection remains one of the most sought after applications for
machine learning in IoT, but machine learning systems often have
difficulty processing data streams. We hope that this paper will
encourage others to use HTM to perform anomaly detection on a variety of
data streams, as well as evaluate other anomaly detection methods using
Thank you for continuing to follow Numenta. We’d love to hear your
thoughts on our latest papers. Tag us on social media or join the
discussions on our HTM Forum.