On reflection regarding your questions on HTM I would point out that the work we are doing here is all experimental, subject to change on discoveries of the research team or any of the member community.
These are still early days in HTM and it has already gone through some pretty significant changes in focus and I fully expect more to change as research continues.
One of the reasons I referred you to biology is that we are still mining that for guidance in research directions.
The closest anyone has gotten to monetizing some aspect of HTM is cortical.io and they are only using bits and pieces of HTM theory.
Reading your introduction I see that you are supportive of startups. These generally have some idea that is proven and ready to be developed for the market. What we are doing here is way further up the development stream and may literally be decades away from commercialization. We all hope that useable results won’t take that long but I would not make book on that; you can lose a lot of money that way.
You are certainly welcome to figure it all out with the rest of us but don’t expect that anyone here can tell you how to fit HTM into a usable application; we don’t know either.
On reflection regarding your questions on HTM I would point out that the work we are doing here is all experimental, subject to change on discoveries of the research team or any of the member community.
One of the reasons I referred you to biology is that we are still mining that for guidance in research directions.
I get that, but I’m an engineer, not a research scientist. I wait for others to do the lab work and papers, and then I look for ways to turn that work into things that are actually useful. I don’t need a full picture, just bits that work. Just the SDR on its own is a major revelation, now what can we do with it?
The closest anyone has gotten to monetizing some aspect of HTM is cortical.io and they are only using bits and pieces of HTM theory.
Thanks for the link. Traditional AI/ML has serious limitations, and this is the kind of use for HTM ideas I’m looking for.
Reading your introduction I see that you are supportive of startups. These generally have some idea that is proven and ready to be developed for the market. What we are doing here is way further up the development stream and may literally be decades away from commercialization. We all hope that useable results won’t take that long but I would not make book on that; you can lose a lot of money that way.
Received wisdom is that it takes 14 years from lab bench to fully developed product. On Intelligence is 2004, Numenta 2005, HTM from around 2011. I’d be surprised if there isn’t something at least worth investing in within 5 years.
You are certainly welcome to figure it all out with the rest of us but don’t expect that anyone here can tell you how to fit HTM into a usable application; we don’t know either.
No, that bit is my playground. After the science is done and dusted.
You might also find this podcast interesting if you haven’t listened to it yet. In it, Francisco (founder of Cortical IO) talks about how he came to the idea of encoding words, and mentions some other interesting ideas besides NLP that the concept of semantic folding might be applied to. If I were looking to monitize concepts from HTM, this is probably the area that I would focus (probably trying to incorporate the idea of distal signals for generating context)