I was re-reading my reply to Chris Chatman's post about biologically accurate modeling and realized that I was focusing on the wrong aspect of his message. Chris is right. There are so many different features that may need to be modeled to achive the result we want; yet every new feature complicates the model, makes it run slower, and makes it more difficult to analyze the behavior of the model and tweak it further. So, what can we ignore?
- Biologically plausible model(s) for different types of neurons (like the model proposed by Eugene Izhikevich)
- Number of neuron types to model; Can we simply use two types: regular spiking (RS) for all excitatory and fast spiking (FS) for all inhibitory neurons. Probably not.
- Model time step; Is 1ms enough? Some of the types modeled with Izhikevich model require 0.1ms time step.
- Axonal delays
- Synaptic plasticity and restructuring
- Synaptic connectivity
- Synaptic noise
- Synaptic currents
- Brain modularity
- Size of the model; Number of neurons? Number of connections? Ratio of excitatory and inhibitory neurons? Different in different modules?
- Sensory input / motor output; How may different types of sensors? How sensitive/complex? Attention mechanism?
- Developmental aspects
- Emotions
- Various types of neurotransmitters and neuromodulators; neuron geometry and localized effects of some neurotransmitters/modulators
- Real-time/on-line vs. static/off-line
- Sleep
- Dendritic geometry (including dendrite-to-dendrite and axon-to-axon connections)
- Glial and other supporting cells
- Wired-in vs. emergent behavior
- Reward/punishment
Leave a comment