Gotcha. By “dimensionality,” I was referring to the feature space which helps define a model, not neccasarily the model itself.
I feel like “possessing a morphology” would almost be a category unto itself. The way you guys are approaching modelling space leaves room for a lot of flexibility, which is nice. I talked about it briefly in that one policy post of mine, but I feel like you can probably handle most representations through the lens of ‘features, models and movement.’ Ultimatly, I think if you guys really focus on getting Monty’s interactions with “physical space” right, all the other more abstract spaces we can imagine (linguistic space, mathematics, social network space) will kind of just be solved emergently.
Also, I’m not sure if you’ll find this helpful, but there was an paper I was looking into some time back, regarding the Anterior Temporal and Posterior Medial systems and how they might be used to assign value to objects within a given scene. Essentially a framework for memory-guided behaivor. I was going to use it to help inform how that “Salience Mapping” policy of mine from the other post might work, but maybe it would help you guys in brainstormning your approach to unsupervised learning? Anyways, if you’re interested:
Out of curiousity, how far into implementing unsupervised learning are you?