I've done a pretty silly thing. In an attempt to learn the fundamentals of sensorimotor inference, I started a project to create a sensorimotor inference engine from scratch all on my own.
My guiding principle was to do everything as simply as possible and see what kinds of problems I run into then, as simply as possible try to overcome those problems.
Though (not surprisingly), this project yielded no substantive code and no unique insights, I do think that my experience has an interesting story. I wrote about the experience at maestroai.com and I wanted to post it here for feedback or to tickle the sensorimotor inference fancy of any of my fellow HTM enthusiasts. It exemplifies the most basic and fundamental challenges of sensorimotor inference and, really, all types of machine learning in general.
Thanks for your feedback!