I think you're confusing layers and regions. Hierarchical regions learn/store different levels of patterns, so a deeper hierarchy might lead to formation of higher level patterns (and thus more "intelligence"). Layers, on the other hand, at least according to HTM, seem to represent different brain functions, such as pattern learning in layers 2/3, behavior generation in layers 4/5, and attention in layer 6 (that's how I understood the sensorimotor video). Increasing the number of layers in neocortex makes sense if you want to develop a new brain function, for example, conscious regulating of some bodily functions, which previously could not be regulated by neocortex, or perhaps as some kind of a brain-computer interface.
Also, intelligence seems to be strongly defined by the number of active connections within/between existing regions. For example, your learning capacity is largest at early age, when the number of such connections is the greatest. On Tue, Feb 10, 2015 at 10:40 PM, Valtér Hégér <[email protected]> wrote: > I believe that a system with significantly more neurons will outperform > one with fewer. Yes, clustering of neurons in certain regions is going to > result in better performance. > Given a limited number (10^8) of neurons, would a model with 5 layers be > able to detect and identify patterns better than one with just four? Is > there some sort of optimal advantage [pattern search space] with just 4 > layers? >
