TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

Upside Briefing: Numenta

Numenta reverse engineers the neocortex for computation and AI learning.

• By Martin Pacino
• November 18, 2016

Name of Company/Solution: Numenta – Hierarchical Temporal Memory

Location: Redwood City, CA

TDWI Product Market Categories: Artificial Intelligence, Predictive Analytics

Product Line Vision: To apply the learning principles of the human neocortex to software and make them available on an open source platform.

Briefing Notes: Numenta is a software company working to reverse engineer the human neocortex. I recently met with Christy Maver, director of marketing at Numenta, to get an update about what they are doing in the areas of artificial intelligence and predictive analytics.

The researchers and engineers at Numenta believe that nature has provided us with a fantastic model for how machines should consume, process, and interpret data -- the human neocortex -- and they are implementing this concept in software on an open source platform. Numenta is using a biological approach to learning based on how humans learn. A large part of our discussion centered on how Numenta is using that process to create a framework for biological and machine intelligence.

"We start with the brain," Maver says. "The brain is the only thing that everyone can agree is an example of an intelligent system. We don't just use the brain as inspiration [as many other machine and deep-learning techniques do]. We constrain ourselves to the brain and use it as a road map because we think it's the most efficient path to machine intelligence."

This unique approach was first envisioned by Jeff Hawkins, who founded Numenta in early 2005. "Intelligent machines will radically transform our world in the 21st century, similar to how computers transformed our world in the 20th century." Hawkins wrote these words in a March, 2015 article for Recode.com entitled "The Terminator is Not Coming. The Future Will Thank Us."

Contrary to recent statements cautioning against artificial intelligence (AI) and machine learning from high-profile thinkers such as Stephen Hawking and Elon Musk, Hawkins' belief is that advances in machine intelligence can be beneficial to society and, indeed, thrilling.

Numenta's approach for harnessing the power of the human neocortex is called Hierarchical Temporal Memory (HTM), which is a theoretical framework for biological and machine intelligence. According to Maver, "HTM is continuously learning and constantly predicting sequences. The key differentiators are: continuous, unsupervised learning, and the fact that it works on a set of universal, rather than task-specific, algorithms, which means it's fully automated -- no tuning is required."

Although the full scope of uses of HTM and cortical machine learning remain somewhat speculative, Numenta has already built example applications to show how the technology can be used in specific disciplines. Here are three examples:

• Financial services: "HTM for Stocks" is Numenta's application where HTM tracks the stock volume, stock price, and Twitter volume for publicly traded stocks and identifies anomalies. The app is designed to display, in real time, the list of publicly traded stocks from those showing the most anomalous behavior to the least. Detecting stocks that are showing a high level of unusual activity and providing a link to the Twitter traffic for each stock allows users to analyze stock performance and start to infer some cause for that activity.


• Human Behavior: HTM has been applied to predict anomalies in the behavior and performance of humans. Tracking metrics such as keystrokes, file access, CPU usage, and app access, the HTM algorithms model each employee individually and can identify when someone does something different or unusual, based on their previous behavior.


• Geospatial tracking: Using speed, position, and direction as variables for a body traveling through space, HTM is being used to root out anomalous movements, too. Deviations from an expected path are identified, allowing monitors to detect problems with how fast something is moving, where it is, or where it is going. This technology is already being employed by harbormasters to monitor and direct boat traffic.

First impressions: The future of HTM is still unwritten. Numenta expects the next stage of development will be in sensorimotor inference, and one day you may just have a conversation with a walking, talking machine powered by an HTM system -- or perhaps HTM-powered systems will be exploring Mars. This is exciting stuff. To be clear, though, this effort is not meant for self-replication of humans. It is meant to understand intelligence and apply principles of intelligence to software.

One aspect I find particularly interesting is that everything Numenta develops is done in an open source environment and users are free to use it as they see fit under an AGPLv3 (open source) license. Companies and individuals are under no obligation to divulge how they use the technology until they choose to distribute a product or service that uses it, in which case they must either make their code available under the same AGPLv3 license or purchase a commercial license from Numenta.

Numenta and HTM may signal the dawning of a new day in predictive analytics and AI.