You're raising a provocative question. I'm surprised--or maybe not
surprised--that this hasn't yet received comments.
Model reduction is a very broad topic. One could say that a form of model
reduction happens every time someone cooks up a conceptual model that
can be mapped into a computational model (because most of our conceptual
models of biological systems have many components that are either too
poorly constrained, or not relevant to a particular research question). The
very act of selecting what a computationally-testable conceptual model
should omit, what it should include, and how much detail to preserve in
that which is included, is model reduction.
But you probably mean "given a computational model, how do I reduce its
complexity?" Typically this involves replacing mechanistic ("works the
same way") representations with functional ("has a similar effect")
representations. The hope is to preserve essential qualitative similarities
(whatever that means--what is the basis for deciding that something is
essential, and how does one judge the degree of qualitative similarity?).
The broad categories of interest are anatomical and biophysical, and
the latter fall into electrical and chemical phenomena/mechanisms.
There is no single strategy that works in all cases. Successful practitioners
accumulate a bag of tricks that sometimes produce satisfactory results.
It also helps to be able to write and/or speak very convincingly.
In any case, I for one would be quite interested in a general discussion
on this topic, here in the Forum and also in a session at the NEURON