'Noise in Cognition: Bug or Feature?' https://psyarxiv.com/438nd/
"First, both perceptual and preferential judgments show that sensory and motor noise may only play minor roles, with most noise arising in the cognitive computations."
"...we propose that the brain approximates probabilistic inference with a local sampling algorithm, one that uses randomness to drive its exploration of alternative hypotheses."
Sort of obvious? But good to see it out there.
Which mechanisms for neural exploration/sampling do you think of as most likely?
1. Periodic and quasi-periodic sweeps
2. 'Real' random perturbations
3. Pseudo-random and/or chaotic pattern generators
4. Deterministic, non-periodic 'principled' exploration
I suspect that the answer is 'all of the above'.
Am I missing any plausible mechanisms?
@DrYohanJohn 1 and 3 in combined approach. 2 is undoable and 4 is too costly.
I think quantum effects at the level of vesicle release can lead to 'real' random fluctuations, so fine balance between excitation and inhibition could be used to 'amplify' synaptic noise. So #2 could be doable.
@DrYohanJohn @vickerse That's an interesting idea. I'm a synaptic physiologist, and I've always wondered about the fact that the random release probability of vesicles is much higher at inhibitory synapses compared to excitatory synapses. Your comment makes me wonder if that's related to keeping the 'balance'? Certainly, there are less inhibitory synapses than excitatory ones. What are the predictions, if there is a change in that balance?
@DrYohanJohn @vickerse
I don't know if there's a specific paper that addresses it, but the data is there across multiple papers.
We've seen in V1 L2/3 pyramidal neurons that average mEPSC frequency is ~5 Hz while average mIPSC frequency is ~10 Hz. This may not seem so different, but if you consider that inhibitory synapses probably are about 10~20% of total synapses on these neurons, it does make the difference in spontaneous release quite high for individual inhibitory synapses.
I've seen similar difference in mEPSC and mIPSC frequency in CA1 neurons as well as in thalamic neurons.
