Neurons

I'm really interested in neural models. I mean models of neurons for doing what neurons do, not connectionist systems.
The Hopfield model is reasonable as it uses the McCullouch Pitts integrate and fire neuron.
The topology is wrong as the brain is not bi-directionally connected nor well connected.
That's an integrate and fire neuron which can be extended to leak, and to my model the fatiguing leaky integrate and fire neuron.
We actually use this for some categorisation tasks.
We also use the neurons to build an agent that includes a natural language parser.
It's cutting edge neuropsychological modelling, but not really great machine learning.
Compartmental models (e.g. Hodgkin Huxley) are more accurate models of neurons.
The hope is that we'll be able to develop a more sophisticated system that is capable of full-fledged AI tasks, for example a data analysis assistant.