I’m loving this video series, thank you for putting these together. I’m on episode 7 (Spacial Pooling). Unfortunately, with life being crazy busy, by the time I get to a new episode, quite a bit of the previous information has been lost.
Is there, or could there be, a short written recap of the contents of each episode? Something around a paragraph or two each that hits every major point in the video and any smaller key bits. That would greatly help in consolidating the information.
If this exists, please point the way!
If not, is anyone interested in putting something together?
(Of course, if others feel this would be a benefit but no one can do it, I will, but like I said, time, etc., etc.)
Also, I am working on Building HTM Systems (WIP Document). Here, I plan on covering all the same material as HTM School, but with interactive visualizations and code examples. I have just gotten back to working on it.
Sorry so say this, but I’m not enjoying the videos. No complaints about the presenter, the production quality is high and the visualisations are useful. Well worth doing, but not what I really need.
I’m having great difficulty recalling the content without reference to a matching set of documents. The videos frequently introduce a new word or concept, explain it (or not), and then assume total recall. Case in point: there was a mention of “boost” and it flashed on the screen, but what did it mean exactly? I forget.
Worse, I have a strong background in both IT and neuroscience, and I know all the words you are using but I don’t understand how you are using them. I know proximal and distal like mother’s milk, but you give them some special meaning that I just don’t get.
So my plea is for a summary of the terms, concepts, links to visualisations and to extra reading. Visualisations and videos are eye-catching and valuable, but words, text and careful explanation of key topics are what I really need.
I experienced some of the same (minor) frustrations at first (now about a year ago), but I realised the shortcomings were due to myself. It is tempting to believe HTM theory is easy, simply because @rhyolight distilled it in single-servings videos. Don’t be fooled: this is really hard material.
So I advise you to aproach this from different angles:
These are not questions the researchers will answer. They’re way too busy for that. But we the enthusiast viewers can try and help. And asking these questions helps us too.
Also don’t forget how generous this endeavor is. Google may organise the occasional talk, and Open AI may publish a few papers once in a while, but noone invests this much energy in reaching out to the community as Numenta does.
Clearly my situation is different. I’m having no problem with the concepts and the basics of the theory seem straightforward. I certainly don’t expect to spend a year understanding what I’ve seen so far.
My problem is mostly about terminology, where common words are used with domain-specific meanings. But a summary of learnings from each video would help too, in the form of text, tables, diagrams for example. I’m sure I’m not the only one who would benefit.
Thanks for that, but a 15 minute video and/or a multi-page research paper are hard work if all I want is to find out: what did you mean by “capacity” or “mini-column”?
To rub salt in (paper to answer a question) open wound - mini-columns:
Please see page 718 in this paper for a nice graphics showing the mini-columns. (23 um spacing between columns)
If you are not familiar with this paper I do recommend it for good grounding in the basic neurobiology underpinning much of what we do here.
Throughout the rest of the paper it shows how the mini-columns are combined in the various sensory modalities. This is discussed in detail in the section labeled “The columnar organization of heterotypical cortical areas.” These larger column structures (300-500 um) seem to be defined by the peripheral receptive field. On some areas this is a particular patch of skin, in others, an area of the retina.
I am familiar with the basic ontogenetic development plan as described by the paper but how that ends up forming the receptive fields in figure 10 blows my mind.
Seriously,
It is extraordinarily hard to make material that addresses all levels of audience, especially, if you are already steeped in the tech you are trying to present.
As a teacher I have dealt with this many times - I can’t remember what is hard for the students as it is all easy for me. Furthermore, every student stumbles on different parts of the material.
In a live setting you gain the tremendous advantage of having feedback from the students as you present the material. If you are a good teacher there is some testing (on multiple time scales) of the students as you go along to see if what you are doing is working. This can be body language, questions, quizzes, worksheets, or tests. In a canned presentation you are flying blind.
When I was doing this for a living I would note the questions as they came up in the curriculum outline and use this to revise the presentation materials.
I did have the freedom to revise the course as this was part of the reason the tech college hired me - to update a stale course. I was told by friends who were teachers at competitive institutions that this is not always an option. Pity.
Sidebars and boxes, lists of terms & definitions in papers, text overlays with key words & definitions in videos, intros and summaries in presentations; these are all good tools but not everyone has experience using them.
Thank you @falco and @Bitking. All that being said, I want the feedback, even if it is critical. @david.pfx is bringing up good points and telling us exactly what he thinks would help. A lexicon of HTM terms like this one might be a good idea to officially maintain. I will think about ways to improve the material.
Great paper, but the minicolumns in the videos appear to be 4 little boxes with dots in them. I get the neuroscience, but it’s the terminology as used in the videos I need definitions for.
Couldn’t agree more. That’s why I’m posting, to help and encourage that process. This is my take on what will help make the on-boarding experience better for newcomers.
Excellent. I was wondering about a wiki, and this does indeed answer quite a few of my questions. Definitely worth linking to (and keeping up to date).
The other link I looked for was how to get started with using the software (on Windows). So far I’ve tried 4 repos, 6 sets of instructions, 2 versions of python, 2 C compilers, 3 build systems and I still don’t have anything that actually works. Any suggestions?
We never officially supported Windows builds. My only suggestion is to look into the “ci” folders of the repos you are working with for instructions for AppVeyor (the Windows CI tool).
Thank you!