Thomas Nowotny and Dan Goodman (not to be blamed for eventual errors in this synopsis as it was written by Michael Schmuker)
Provide Brian to a high-performance GPU-enabled backend using GeNN. Brian has already such a backend in form of the NeMo backend, but GeNN complements NeMo's capabilities for modular network architectures and dense connections. In addition, GeNN supports neuronal models which NeMo doesn't.
We have identified the architecture for the interface which is required to connect Brian and GeNN. The interface is currently being written. It uses SWIG for making the GeNN API available under Python. On the other end, we'll use brian's code generation facilities to provide GeNN with user-defined neuronal and synaptic models.
We'll implement the brian interface. Once this is finished, that interface will probably make the PyNN → GeNN interface obsolete, as brian already has a pyNN interface. One would then only have to modify brian's PyNN interface to provide an option to use the GeNN backend.