Numenta Newsletter November 2016

Numenta Newsletter — November 10, 2016

I’m pleased to share that we have published a new research paper titled
Continuous Online Sequence Learning with an Unsupervised Neural Network Model,
in the peer-reviewed journal Neural Computation. This paper is a companion
piece to a paper we published earlier this year that introduced Numenta’s theory
of how the brain learns and understands sequences of patterns. We refer to the
theory as HTM sequence memory. In this new paper, authors Yuwei Cui, Subutai
Ahmad and Jeff Hawkins compare HTM sequence memory to other state of the art
machine learning methods for streaming data.

Writing this paper is another important step for Numenta, as we continue to
publish peer-reviewed research that demonstrates the importance of brain theory
in creating machine intelligence. What makes this paper particularly interesting
is that it compares HTM sequence memory to several other sequence learning
algorithms, including statistical methods and recurrent neural networks. HTM
achieves comparable accuracy results while displaying several important
properties that apply to both biological systems and real-world streaming
applications. These properties include continuous online learning, the ability
to make multiple simultaneous predictions, robustness to sensor noise and fault
tolerance, and good performance without the need for task-specific tuning. We
are finding an increased interest in brain models from the machine learning
community. We hope the performance comparisons to a wide variety of techniques
in this paper will introduce a neocortical theory-based approach to a new
audience and encourage people to apply HTM to real-time sequence learning
problems.

Speaking of machine learning, Numenta will be sponsoring MLconf SF on
November 11 for the second year in a row. MLconf targets the San Francisco
machine learning and data scientist community. At our booth, we will give demos
of our tools and example applications, discuss our latest papers, and answer
questions. Please stop by if you’re planning to attend. If you’re not in the
area, you can still check out this recent blog post from MLconf Technical
Chair, Nick Vasiloglou, who interviewed Numenta engineer Austin Marshall about
Numenta’s view of neural networks and AI.

In case you missed them on our Twitter feed, we have some new pieces on the
website, including a preprint of a journal submission titled
“The HTM Spatial Pooler: A Neocortical Algorithm for Online Sparse Distributed Coding”
and a guest post for InsideBigData on
“The Importance of Brain Theory in Machine Intelligence”.