Whether or not you accept the Bayesian framework of the brain wholesale there are good reasons to expect that most of its microscale processes are probabilistic. An interesting article by Kasabov proposes a spiking neural network model to account for this while maintaining biological consistency. I can’t explain the math easily, but here are a few of the biological underpinnings of his action potential models:
- Fast excitation post synaptic potentials are mediated by AMPA receptors
- Slow excitation post synaptic potentials are mediated by NMDA receptors
- Fast inhibition post synaptic potentials are mediated by GABA-A receptors
- Slow inhibition post synaptic potentials are mediated by GABA-B receptors
The goal of nearly all of these types of models is to fundamentally modify the parameters during the learning phase to achieve faster and more accurate predictions during the testing phase.
Kasabov N. 2009 To spike or not to spike: A probabilistic spiking neuron model. Neural Networks, In Press, doi:10.1016/j.neunet.2009.08.010.