Introduction to Numenta's Research Release from Jeff Hawkins
To our developer partners:
Although you may be anxious to download Numenta's software and give it a try, you should read this brief note first. I will give an introduction to what this release is for and what you should expect as you use it.
Why did we create Numenta?
Numenta was formed to develop and promote a technology called Hierarchical Temporal Memory, or HTM. We are confident that the principles underlying HTM are the same principles that govern much of the operation of the human neocortex; thus, HTM enables the creation of machines that have some of the capabilities of the human brain. We see HTM as a fundamental new computing methodology able to solve longstanding problems in artificial intelligence and machine learning. Be sure to read the white paper on our web site that describes the basics of HTM and what it can do.
We believe that the potential applications for HTM are broad and far-reaching; yet we know that our small company can only work on a piece of this application potential. Consequently, our goal is to create an industry built on HTM, opening up the platform to the world of creative developers. To this end, we have created a set of software tools that allows anyone to experiment with HTM, to apply HTMs to different problems, and to extend the tools. We call this set of tools the Numenta Platform for Intelligent Computing or NuPIC, and we have made it available for anyone to download under license.
What does Numenta hope to accomplish with this release?
Our immediate goal is to get more people working with us to explore and extend HTMs, and to build a community of HTM pioneers. To further this goal, we are making our tools available free of charge. We have put in place on-line forums and other support mechanisms to help our community of HTM developers share insights and information. We hope that after reading this note, and viewing the other material on Numenta's website, you will be enticed to learn more about HTM and to start experimenting with NuPIC.
Why are we calling this release a "research release"?
We chose to call our first release a "research release" for two reasons.
1) HTMs are new.
We have found that it takes some developers months to get familiar with the concepts, and maybe a year to get commercially useful results. It also is possible that your first attempts at using HTM will be valuable for learning, but not valuable for commercial applications. The learning curve is similar to the learning curve for writing computer software. It takes anywhere from several months to a year to become proficient in writing computer software; HTMs have similar complexity.
2) Our platform, NuPIC, is immature.
We also call this a "research release" because the release has some shortcomings. To begin with, NuPIC includes a set of embedded learning algorithms. The first implementation of these algorithms can solve only a subset of the problems that theoretically can be solved by HTM. For example, although the first release can solve many pattern recognition problems it lacks the ability to make predictions forward in time or to recognize patterns that are inherently time-based such as language. We believe we will make progress during 2007 addressing these and other issues. We invite developers also to contribute to algorithms development, and have offered source code to these algorithms to facilitate such work. In addition to the maturity of the algorithms, the platform has a few parameters that need to be set by experimentation and analysis, which makes it harder to use today than it ultimately will be. We will be automating these parameters and making our tools easier to use over the coming year.
Because of these issues, those developers using the first release of NuPIC should consider themselves as HTM pioneers, researching the capabilities of this new technology.
Why are we making NuPIC available now?
We have been contacted by dozens of researchers and scientists who are excited about HTM and by our work at Numenta. These people are anxious to work on HTM, are willing to be pioneers, and are willing to accept the uncertainty associated with a new technology. We are making our tools available so that these sophisticated developers can start building a community around HTM technology. NuPIC has been under development for 18 months, is pretty solid, and is well documented - including several examples to make it easy to get started - so we're ready to open up to more developers, even while knowing that we do not yet have benchmarking data, and we cannot make guarantees about applicability to specific problems.
If you are looking for a more turnkey solution, you might want to wait until the platform has matured.
Why should you invest your time to learn the Numenta platform?
We believe that HTM will turn into a powerful computing paradigm with applications in nearly every industry. Our confidence comes from the facts that HTM is theoretically elegant, has a strong biological mapping, and is working well in our testing so far. If we are right, then HTM provides the opportunity to build many successful businesses and products.
There are opportunities for applying HTMs to products, creating services with HTMs, creating custom hardware for HTMs, being an HTM consultant, selling pre-trained HTMs, and creating new HTM tools.
It is not often that one gets the opportunity to get in on the ground floor of a new technology with such broad reaching potential. If you find HTM exciting, if you believe the concepts underlying HTM are useful, and if you are willing to accept the risks as well as the rewards of being an early adopter, then we invite you to download the NuPIC tool set and get started. We will support you the best we can, and we welcome the opportunity to work together.
Resources
The HTM white paper can be found at http://www.numenta.com/Numenta_HTM_Concepts.pdf.
