NuPIC aims at imitating the brain to take Machine Learning further

Sept. 5, 2017, 3:11 p.m. By: Prakarsh Saxena

NuPIC

Numenta, the machine intelligence company founded in 2005, is the father company of NuPIC- Numenta Platform for Intelligent Computing. NuPIC is a machine intelligence platform that implements the HTM Learning Algorithms, and is an implementation of the computational theory of the neocortex. At its core are time based learning algorithms that store and recall temporal patterns and learn from them. The HTM Technology is suited to address a myriad of real world problems, particularly those with the characteristics of streaming data, underlying patterns in data which change with time, time based patterns etc.

What is HTM Technology?

HTM Technology forms the crux of the NuPIC tool. It stands for Hierarchical Temporal Memory and is a biologically constrained theory of machine intelligence originally described in 2004 by authors Jeff Hawkins and Sandra Blakeslee in their book On Intelligence. HTM is based on neuroscience and the physiology and interaction of pyramidal neurons in the neocortex of the human brain.

A typical HTM network is basically a tree- shaped hierarchy of levels that are composed of small elements called nodes. A single level in the hierarchy is called a region. Each HTM region learns by identifying and memorizing spatial patterns- combinations of input bits that often occur at the same time. It then tries to identify temporal sequences of patterns that are like to occur after one other. An HTM Model Neuron also tries to learn by modeling growth of new synapses, instead of focusing on change of weights of synapses.

HTM are known for their robust nature to store, learn, infer and recall high order sequences just like a normal brain does. But unlike other machine learning models, HTM has the ability to learn time based patterns in unlabeled data on a continuous or real time basis. The algorithm is well suited for applications such as prediction, anomaly detection, classification and sensorimotor applications due to its extremely low susceptibility to noise and high capacity to learn patterns.

In short, HTM is basically an understanding of the logical neocortex system of our brain as close as it can get, and is hence an open source project which invites updates for the betterment of the algorithm functioning.

Where can NuPIC be used?

NuPIC has found a lot of use in the Business Modelling, IT Applications and Analytics sector of the industry. It can be easily installed on Linux/ OS X 10.9 or higher/ Windows 64 bit. It although has some dependencies on the following:

  • Python 2.7

  • pip>=8.1.2

  • setuptools>=25.2.0

  • wheel>=0.29.0

  • numpy

  • C++ 11 compiler like gcc (4.8+) or clang

Additional OS X requirements:

  • Xcode command line tools

NuPIC is available on GitHub for explorations and use. You can see the source code here, or visit the organization at numenta.org.