Numenta Anomaly Benchmark (NAB), and Numenta HTM Challenge

Donna Dubinsky • CEO

Numenta Newsletter October 2015

Numenta Anomaly Benchmark (NAB), and Numenta HTM Challenge

To Numenta newsletter subscribers:

I’m writing with a few exciting updates from Numenta.

To start with, I’m pleased to announce that we have a paper accepted on the
Numenta Anomaly Benchmark (NAB). I’ve written to you about the development of
NAB in the past. NAB is essentially two things. First, it consists of over 50
streaming data files with anomalies that are marked for ground truth. These
files are a combination of real-world data sets along with some simulated data
sets. Second, there is a scoring mechanism that is designed for streaming data.
Numenta’s HTM algorithms do a great job at finding temporal structure, learning
continuously, and detecting anomalies early. These are valuable properties for
real-time streaming applications but benchmarks designed for batch data don’t
incorporate them. For example, traditional benchmarks do not give “credit” for
finding an anomaly sooner rather than later. Our development of NAB is an
effort to create such a benchmark.

We will be releasing NAB into open source such that the community can add new
data sets, propose different scoring mechanisms, and test/compare other
algorithms with HTM. We hope that the developers in the community will get
behind NAB and enable a robust comparison of our anomaly detection with other
techniques.

For more information on NAB, go here. We will
be presenting NAB at two conferences
(MLconf SF on November 13 in
San Francisco, and the
IEEE Conference on Machine Learning and Applications
on December 9-11 in Miami), so if you are in the vicinity, we hope you will be
able to attend. Of course, once the paper is published, we will send you
the link.

My second topic is an update from the NuPIC community. We recently announced
the HTM Challenge. The idea of the HTM Challenge is to provide a structure to
encourage members of our open source community to create HTM sample
applications. And, to add an extra incentive, we’re offering cash prizes of
$3000 for first place, $2000 for second place and $1000 for third place, along
with some fun door prizes.

The Numenta HTM Challenge is an online
contest open for submissions now and running until November 14th. Participants
propose ideas for real-world applications of HTM technology and submit them for
approval before starting work. We will check that the problem being solved is
applicable to HTM, and that the data being analyzed is a good fit for HTM.
Once a project has been approved, hackers have until November 6th to work on it
and submit a demo video for judging.

This Challenge can be completed entirely online, but we encourage you to try to
attend the onsite event that
culminates the Challenge on Saturday, November 14 in Redwood City, California.
At this event, all demonstration videos will be viewed by a judging panel and a
live audience. Judges will get a chance to comment and ask questions to
submission authors (either live or over the phone). At the end of the event,
we’ll announce the winners and cash prizes will be awarded.

In addition to judging, there will also be at least two educational sessions by
Numenta team members Subutai Ahmad and Jeff Hawkins. They will be talking about
the history and evolution of HTM algorithms, the Numenta Anomaly Benchmark, and
some details about new algorithm development.

You don’t have to participate in the Challenge to attend the onsite event.
Maybe you just want to come see the latest presentations from Numenta or watch
the Challenge demonstration judging. That’s great! We welcome you to mingle with
us HTM enthusiasts.
Register here.

And as long as you’ll be in the area, there is a community event being planned
on Friday, November 13, the day before the onsite event. I hope you’ll come meet
the rest of the HTM community. This event is run by the NuPIC community, for the
NuPIC community. A schedule is in the works.
RSVP here.

Stay tuned for more information on NAB and on the HTM Challenge. I’ll update
you as both projects develop.

Lastly, let me remind you that Avik Partners has now taken over the Grok for IT
product. They have enhanced the product and are now in
beta. If you want to experience breakthrough anomaly
detection on IT infrastructure, register to give it a try.

Donna Dubinsky • CEO

Leave a Reply

Your email address will not be published. Required fields are marked *