HTM Theory for Newbies

When I started out with HTM theory I was having a lot of trouble understanding the concepts as it was the first time I was working on anything related to Artificial Intelligence. However with time and patience I slowly developed a basic understanding of the theory. I have written an article in my blog that might help you understand the concepts in an easier fashion. The article is basically a collection of all the papers and the videos I had seen, combined and written in a way so that a even a tenth grader can understand it. If you are interested please follow this link.
Please do let me know if I have made any mistake or if you think there is a better way the article could have been written. Hope it helps. Thank You. :slight_smile:


I love the drawings you made! Good stuff!

I only have one correction: replace “Temporal Pooler” with “Temporal Memory”. When we talk about temporal pooling, we’re talking about a different thing that the current temporal memory algorithm you describe does.

I will share your article, thanks for writing it. :smiley:


I have made the corrections in the article. I am glad that you liked it. I am just doing my part to help the community grow and be better. And thanks to you for all your hard work and dedication. :grinning:


Hi Kaustab, your website will be a great resource as you learn everything behind the robotics and AI field. I don’t have any suggestions for your article but I’d invite you to make some business cards with a link to your site. When you meet with people and talk about all these exciting ideas they will love having something physical to go along with the discussion and remind them of you. I own a site with a similar mission and helps get the word out.


1 Like

Your article helped me where I am still struggling to understand the HTM theory. Today on 1 June 2017 I created a henceforth monthly “Numenta” folder on my computer to gather up an accretion of HTM details. In your blog-post (where I left a separate comment), you say that “when an HTM model sees an input for the first time, it doesn’t know the context of that input, so it activates all the cells in the activated column to represent that input and then randomly assigns a cell to represent that input for future reference.” This idea of randomly assigning a cell sounds similar to concept-formation in my How the Mind Works theory, where there is no single Grandmother Cell for each concept, but rather randomly assigned cells flesh out the concept to make it redundant and persistent over time.

1 Like

That’s a great idea man but I am just starting out with my blog. The goal is to spread awareness about AI amongst everyone and you can really help me achieve that by spreading the word. As far as business cards go, I do hope to make them one day but only when I have achieved a bit of momentum with my site. Thanx for the idea and keep up the good work. :slight_smile:

1 Like

I was struggling myself at first. But do take your time and understand the very basics at first before moving forward. I am glad I could be of help. You can always clear your doubts by asking about them in the forum though.

This means a lot to me. Thanks a lot. :grinning: