Hey Neouro Science lovers , I have question for the nroader community on Sensorimotor Theory of Numenta :
From what i know layer 4 is the predominant input layer into the cortex which received the feed forward input connections from thalamus and the allocentric location as well of whatever object is being sensed atm. But Sensorimotor theory states that object recognition and further prediction post Object recognition happens in layer 2/3. and layer 2/3 does not have any incoming allocentric location coming to the layer, then how does layer 2/3 know what is the orientation of the object in 3D space for it to make future depolarizations of the input layer for the object?
2 Likes
Iām recently of the opinion (that I will hopefully be able to demonstrate someday), that L2/3 maintains a generative model of the input it expects to receive from L4. So long as enough of the incoming activations are anticipated by the generative model, our perception is that of normality. However, when portions of these models diverge from observations, then the mismatch triggers an attention mechanism which drives information seeking behavior in order to obtain more input for training the model to anticipate such situations in the future.
So, to answer your question, the 3D aspect is part of a generative model, much like NeRF, or Gaussian Splatting.
4 Likes