Numenta is creating a new type of computing technology modeled on the structure and operation of the neocortex. The technology is called Hierarchical Temporal Memory, or HTM, and is applicable to a broad class of problems from machine vision, to fraud detection, to semantic analysis of text. HTM is based on a theory of neocortex first described in the book On Intelligence by Numenta co-founder Jeff Hawkins, and subsequently turned into a mathematical form by Numenta co-founder Dileep George.
Numenta is a technology platform provider rather than an application developer. We work with developers and partners to configure and adapt HTM systems to solve a wide range of problems.
HTM technology has the potential to solve many difficult problems in machine learning, inference, and prediction. Some of the application areas we are exploring with our customers include recognizing objects in photos, recognizing behaviors in videos, identifying the gender of a speaker, predicting traffic patterns, doing optical character recognition on messy text, evaluating medical images, and predicting click through patterns on the web. The world is becoming awash with data of all types, whether numeric, video, text, images or audio, making it challenging for humans to sort through it and find what’s important. HTM technology offers the promise of making sense of all that data.
An HTM system is not programmed in the traditional sense; instead it is trained. Sensory data is applied to the bottom of the hierarchy of an HTM system and the HTM automatically discovers the underlying patterns in the sensory input. HTMs learn what objects or movements are in the world and how to recognize them, just as a child learns to identify new objects.
Numenta's first implementation of HTM technology is a software platform called NuPIC, the Numenta Platform for Intelligent Computing, which is available to developers under a free research license. Numenta also is developing a Vision Toolkit and a Prediction Toolkit that will simplify the task of creating HTM networks for specific problems. Interested partners and developers should download NuPIC for experimentation and register for the Numenta Newsletter to learn about future releases of the Toolkits as well as other developments in the HTM world.
Vision4 Demo Application
This demonstration application, called Vision4, gives a sense of how Hierarchical Temporal Memory (HTM) performs in recognizing objects in static images.
At the Numenta HTM workshop in June 2008, we released an example of a NuPIC vision application that was similar to Vision4. We are releasing the Vision4 demo for two reasons. First, we have structured Vision4 as a self contained application, allowing non-technical people to install and use it, unlike the prior release, which required programming skills. Second, we have made improvements to the accuracy of the included HTM network. As a result, Vision4 performs much better than the workshop release.
Vision4 contains an HTM network trained on four image categories. It also includes a set of 50 novel test images and allows you to experiment with your own test images. Although Vision4 only recognizes only four categories of objects, the number of images and possible variations of images within those categories is huge. The Vision4 application isn't perfect but it performs well on what is widely acknowledged to be a difficult pattern recognition task.
No comments:
Post a Comment