How do the neuronal ingredients, synapses, neurons, their electrical and chemical signals and the distributed, interacting networks that they form, represent and process information (compute)? So we accept that there is incoming information (visual, auditory, etc.) which the system utilizes (encoding/decoding) in order to behave. So we can think of 3 main problems that relate to brain or any computational system in particular:
2. What is the algorithm for the problem to be solved?
3. Finally, how do these algorithms are implemented by the various brain regions?
The main mission of the brain is to compute. Now let's see examples of single cells in the behaving brain. The most important example is the Hubel and Weisel one for which they received the Nobel prize. They recorded a particular region of the visual cortex of a cat, implanting an electron and trying to find out what are the parameters being computed by a single cell in the visual cortex. It was found that the cell was firing at a particular occasion and more specifically, when a line was moving.
The most fundamental, direct and early paper on the field, trying to understand a neuron as a micro-chip and computation device is the McCulloch and Pitts (M&P) point-neuron (1943) ("A logical calculus of the ideas immanent in nervous activity"). This paper was inspired by two properties of the neurons: a) the all-or-none property and b) the excitatory and inhibitory nature of synapses.
1. Correct interpretation of experimental results (also provided predictions).
2. Gain insights into key biophysical parameters (enables compact description of the physiological behavior).
3. Suggest possible computational (functional) role for the modeled system.
This theory tries to model mathematically the impact of (remote) dendritic synapses (the input) on the soma/axon (output) region. Rall thought that over-simplification of representation on neurons (like in M&P neuron) may have negative implications (things missing) on total understanding of the system (he insisted on the contrast between the schematic neuron and the real, histological neuron). He discovered that when injecting current to a cell soma, most of this current flows to the dendrites (away from the soma) instead of flowing through the membrane.
So in a real neuron, soma is not isopotential, in anatomical, physiological terms. This means also a) the dendrites are not isopotential electrical devices, b) the voltage attenuates from synapse to soma, c) it takes time (delay) for the PSP to reach the soma (because we have a distributed electrical system), meaning it takes time to see the effect of the synapse to the cell body and d) that somatic EPSP/IPSP shape is expected to change with synaptic location. Rall looked at a dendritic tree as a set of connected cylinders:
1. Dendrites enable neurons to act as multiple functional subunits (first locally, then globally, at soma).
2. Dendrites can classify inputs.
3. They can compute direction of motion.
4. They can improve sound localization (in the auditory system).
5. They help to sharpen the tuning of cortical neurons making them more accurate.