Computational models of the brain exist at several different levels of description, from the very precise modelling of a unique spiking neuron, taking into account ion channels and/or dendrites spatial geometry, up to the modelling of very large assemblies of neuron that express complex dynamic interactions. I'm interested in a mesoscopic approach of the brain where the computational paradigm is grounded on the notion of a unit that is essentially a set of time dependent values varying under the influence of other units via adaptive weighted connections. The evolution of the units' value and weights are simply defined by a set of differential equations. This is a strongly constrained framework that has been designed such as to avoid any modeling artifact like a central supervisor or a homunculus. If some properties are to emerge from our models, we want to make sure they are properties of the model as opposed to properties of the software that run the simulation.