Hi, I thought it would be interesting to discuss how SDR’s influence the number of patterns we can detect/predict/reconstruct in a single SDR, in this case how two different patterns scents vanilla and lilac when combined as vanilla&lilac are processed and how this influences how many scent patterns can be layered on top of each other before we start seeing them as a new thing.
If for the sake of argument, we assume that lilac and vanilla, are SDR’s with 2% activation and that the nose when it encodes it only sends an SDR of 2% to the brain for processing.
Of the 2% that arrives, 1% is from the vanilla pattern and 1% is from the lilac pattern and both can be predicted and reconstructed by the cortex, such that memories can be retrieved relevant to those scents and a new memory can be encoded of the union.
But there is an exciting implication though, information theory wise. There must be a capacity at which we stop being able to predict the signals in the mixture.
If we assume that we can only reconstruct the SDR signal if we have around 10% of the activations which is 0.2% in this case. Then the max scents in a mixture we should be able to distinguish is 2% / 0.2% = 10 roughly. In practice, it depends on training(because of chunking). but it might be the reason why soft drinks have aroma mixtures of 15-20 ingredients.
To try and verify, we could think about the noses of dogs and how we relate scent mixture prediction relation wise. And this is where I need a little help on the following problem:
Dogs, in general, have roughly 800 chemoreceptor types whereas humans only have 400, dogs also have a nose surface area that is 17 times greater than humans.
But is nose area and receptor count the things that make it easier to predict the patterns in a mixture? or is it the increased amount of activations from a larger nose and larger SDR that makes memory easier to access?
My best guess would be that the SDR activations of a dogs nose would be 34 times greater than humans and so the nerves would transmit that much more activations ( 34*2% ) / 0.4% = 170 different scents that can be picked out from a mixture.
But this seems somewhat naive, thoughts?