Brette and colleagues' implementation of the COBA model.
Well, the error isn't in "Brette and colleague's implementation"--it's in the NEURON implementation, which was done by Michael Hines and me. Our "coba" version of the Vogels & Abbott model has synaptic time constants that are twice as large as they should be. Sorry about that mistake, but thanks for bringing it to my attention. Fortunately it does not affect the qualitative results published in Brette et al. 2007. The fix, of course, is simply to change the assignment statements that you quoted. We will make revised source code available on ModelDB
http://senselab.med.yale.edu/modeldb via accession number 83319.
The "cobahh" and "cuba" implementations have correct time constants. There is no issue with the "cubadv" model, in which the time course of all synaptic transients is governed entirely by the "membrane" time constant. Also, the time constants were correct in the PyNN implementation (which generated NEURON and NEST models that were used for Fig. 24).
Final note: the discovery and correction of this error provide yet another demonstration of the importance of the general principle of sharing model source code, and of the specific value of ModelDB as a mechanism for model sharing.