That’s very interesting idea.
Considering smell is the only sense which reaches cortex bypassing the thalamus makes one asks whether:
- does “smell encoding format” is the primordial one, used by cortex algorithms
- and all other sensory input is “translated” to make their encodings be compatible with whatever these algorithms are
- which could be, at high granularity, like the one described in the article
- and maybe even inner processing up to abstract thinking is based on slight variations of “smell encodings and algorithms” to represent things like distance, direction, similarity, proximity and paths throughout and between all concepts/objects/ideas we can think of
I havent read the linked paper yet.
But just want to point out that when I look at diagrams of the smell-processing circuits of the drosophila brain, I can’t help but see a spatial pooler tied directly to a CLA classifier.
Did a few tests based on the circuits described in the paper, and it sure looks like its doing diffusion on a learned topology.
but I’m having a hard time comprehending how the goal cells can extract any virtual metric other than the distance to the most active point cells
Interesting the kinds of diffusion patterns one can get from this “map network” by changing the correlation between the inputs.
Which paper? The endotaxis one or the mushroom body one?
The endotaxis one.
Ok then regarding:
The direction not only distance can be inferred by sensing variation of magnitude of a “smell” (virtual or real) in the immediate proximity of the agent,
yea, but what I dont get is how the magnitude of a virtual smell can be computed from the autoassociator output.