If you’d like to share the projects you are working on with our community, please respond to this thread. I like to keep the community informed about what I’m currently working on. You are welcome to join in at any time you have something you’d like to share!
Matt’s Current Projects
Interview with a Neuroscientist
I started a new YouTube series where I interview neuroscientists! The first was David Eagleman. I have already recorded two more interviews with Michael Berry and Eric Jonas, which will be published over the next month.
These videos were (and still are) a lot of work to plan and produce. The point is to engage a more general YouTube audience looking for neuroscience education videos, and pull them into the world of HTM. I hope to get more subscribers and grow our YouTube community with this outreach. It is also great to have these scientists validating the approach of Numenta and HTM just by speaking with me on related topics.
Layers and Columns Visualization and Demo
With the help of @lscheinkman (who is doing most of the coding), I’m putting together a demonstration of what is possible when using the cortical circuit architecture defined in our Layers and Columns paper. This demo runs NuPIC and sensorimotor research code in the Unity framework, where a simulated hand will grasp several different objects. Each fingertip is covered in sensors and will provide proximal feedforward input to one cortical column. With 5 columns (one for each finger), I am trying to show the convergence in the columns toward a learned object as touch sensations are processed. I also want to show objects being learned over time by visualizing the output layers converging on common representations for the object.
As part of this, I’ll be illustrating some aspects of neuronal receptivity zones (apical, basal, proximal) as well as emphasizing the idea of receptivity zones in layers as well. I especially want to try to communicate the modular nature of these cortical circuits and the potential for novel exploration of this space.
Upcoming Presentations and Meetups
I’m talking at Strange Loop on Sept 28 and OSDC West in SF on Nov 3 (see https://numenta.com/events/ for details). I’m in the process of updating this presentation to include “Layers and Columns” in addition to the sequence memory explanations I have been giving. I am hoping as the LAC visualizations I defined above are created, I’ll be able to revamp this presentation so that “Layers and Columns” takes a center stage.
Planning out upcoming YouTube videos
In order to gain a respectable YouTube following, you must have a consistent video release schedule. As of last week, I am publishing a new video on the HTM School YouTube channel every other Friday. Here is my tentative schedule:
Thanks for sharing your work. I am especially interested on your demo of sensorimotor concept with Unity LAC. I believe that the interaction with VR will be the best demo…
I’m also very interested in the Unity demo. I have been planning to do a presentation on HTM at the developers guild where I work and wanted to include LAC, but need an engaging way to convey a lot of information within an hour.
At my work, I had to do a series of three presentations before people started to see the possibilities and learn to think in an HTM way. Just encoding and SP is a lot to process, but you don’t see the payoff until you get to TM, TP, or SMI.
I had to do a series of three presentations before people started to see the possibilities and learn to think in an HTM way.
Yeah I had a feeling I was going to need to do this.
If you’re anything like me, you also need time to digest it. I read On Intelligence, subsequent papers and watched HTM school over a period of a couple of months, and definitely needed the in-between time to ponder it and build the link between the theory and the real world.
I honestly did not understand most of HTM theory until I started creating HTM School. There was another BIG aha! moment that came when I understood the implications of Layers and Columns.
The Strange Loop talk (Matt speaking at Strange Loop in St Louis) is the best sub-hour HTM talk I’ve seen so far, I think I’ll borrow very heavily from it when deciding what to include (with credit of course ).
Nope, but @lscheinkman is working on it. We’ll have to figure out how to make it open source once it is done. We have to make sure we’re not sharing any proprietary 3D models.
I think it should be enough for basic experiments when the unity framework outputs the features and location of each finger in a way so that we can receive it and put those information into our experiments. In other words, your framework looks like a black box intuitive generator of features + locations!
I’ve been following the Numenta crowd with spotty interest for at least a decade now, always expecting to one day venture over and catch the first glimpses of a major breakthrough of some kind. No Nobel Prizes or computers mimicking biological cognition on the retail market …, yet, but I’d still keep money in your stocks and bonds were they available
My hunch has always been that you guys need some clues: quickly stated, it’s not the neocortex per se but the evolution aspect which completes the picture. I’ll soon have a more robust, first presentation set up for the world to grasp a more complete picture of my discovery, until then perhaps the people at Numenta would benefit from the information I share now. Numenta researchers would do well to realize the other pieces to the neocortex puzzle.
Matt, I thunk Jeff would freak the moment he fully grasps to realize what this verdant, virgin Cartesian mathematics involves.
I’m a masters student over at the Rochester Institute of Technology in the nanocomputing laboratory. Our group does a lot of work in neuromorphic computing at the hardware level but at the software level we often work with HTM. My background is in mathematics and computer science so my thesis is centered around HTM.
Specifically, I’ve designed some carefully crafted experiments that illustrate the current specific limitations of the anomaly detection algorithm with HTM as described in the 2017 paper “Unsupervised real-time anomaly detection for streaming data.” I’ve derived some theoretical explanations to those limitations as well. I’ve developed a new algorithm that overcomes these limitations based on my observations. Stay tuned!
).