HTM Meetup, New Columns Paper & More
I’m excited to share that our first paper on sensorimotor inference has been accepted and published in the journal Frontiers in Neural Circuits. We posted an earlier version on bioRxiv in July while it was undergoing review, and the final, peer-reviewed version is now available under its new title, “A Theory of How Columns in the Neocortex Enable Learning the Structure of the World.”
Much like our “neuron paper” published in the same journal last year, we think this “columns paper” will be a seminal piece for neuroscience and offer significant contributions to brain theory. While the neuron paper introduced a new understanding of neurons and their internal processing, the columns paper proposes a theory of how layers of neurons work together to learn the structure of the world through movement. It introduces what we refer to as the “missing ingredient” of brain theory, which helps explain things like how we form 3D perceptions from 2D inputs, and how we learn abstract concepts in the same way we learn physical objects.
The timing of this publication is particularly auspicious for us, as there’s an interesting trend developing in the AI field–one that you may have noticed as well in recent media coverage: AI leaders are calling for more neuroscience. Prominent AI researchers have been commenting on the direction of AI and notably, the limitations of current techniques. Geoffrey Hinton, often referred to as the father of deep learning, recently suggested that maybe we need to “throw it all away and start again…I don’t think it’s how the brain works.” Demis Hassabis, co-founder of DeepMind, proposed that it’s worth trying to understand the brain and how it achieves intelligence, as it’s “… the only existing proof we have that the sort of general intelligence we’re trying to build is even possible.” These comments reflect a growing recognition that deep learning, while effective and impressive in many ways, is not going to get us to machine intelligence. Yet the brain offers a roadmap that can.
This is something that we have long believed at Numenta. We’ve spent years studying the brain, so that we can understand the core principles of intelligence–principles that cannot be ignored if we want to build intelligent machines. One of those principles is the recent discovery pertaining to sensorimotor inference, which the new paper documents. We believe it resolves many unanswered questions about how the brain works.
Now that the peer-reviewed version has been published, we plan to present our findings through a variety of venues. For those of you in the Bay Area, Jeff will be speaking on Friday Nov 3 at the “HTM Layers and Columns Meetup.” In addition to the paper and its figures, he will discuss the significance of our recent research advances and the importance they will play in AI and cortical theory. Space is limited, so reserve your spot today.
Jeff and the research team will also be visiting neuroscience labs over the coming months to engage with scientists who can experiment with our theories and verify their potential. If you’re interested in scheduling a visit at your institution, contact us here. We may not be able to accept all requests but will consider invitations as schedules permit.
In the meantime, we hope many of you will read the paper and watch the accompanying video. As a reminder, you can find links to our entire library of research papers on our website. Let us know what you think; we’d love to hear from you. Send us a message, tag us on social media, or join a discussion on the HTM Forum. Thank you for your continued interest in Numenta.