I was inspired by @sebjwallace to make another animation of the CLA! This one is an educational animation. At the moment, it only shows the first time step, and this version has a couple hiccups. I’m still working on the right way to present the details of some things e.g. the permanence values of things. The file is a gif, but too large for uploading directly. It is available here:
https://drive.google.com/file/d/0B_jvKu0N6hcaNDNkQzlKZkFoNUk/view?usp=sharing
It should be public to view. Don’t be surprised if it takes a long time to load, it’s a large file.
The animation needs to be explained, still, of course. It does not work on its own. But it shows a lot of the details. The tough part of an animation like this is keeping track of everything, and trying to show the evolution of the algorithm, without losing any information. For example, I’ve found it typically takes 15 time steps of spatial pooler to get a good steady state, and for temporal memory it takes even longer.
The way to tackle it, I figure, is to show just the spatial pooler, and just the temporal memory, and to show them first in the startup state, then in the steady state. Perhaps distinguishing between learning and inference alone, as in the white paper pseudocode. There’s a lot of things to consider, and a lot of data to show. I really can’t process all the stuff that’s going on in @sebjwallace 's animations, at least I couldn’t on the first few viewings, so in case anyone else is in the same boat I hope this helps!
3 Likes