Aaron Russell Voelker

FYI I have a [reference implementation](https://github.com/ctn-waterloo/nengo-temp/blob/voja/nengo/builder.py#L1644).

Tagging myself here because I had an idea for how to fix the fact that this transform introduces additional filtering (and error propagation) with the intermediate population.

Bump; is this the most recent implementation, or has anything else been done since in this direction?

Thanks! So, I finally got around to trying the secret approach I mentioned at the top of this thread. The idea is to implement a custom solver that declares (arbitrarily)...

> You could also pose this as a NNLS problem by making the activities for the inhibitory neurons negative, and having all positive weights. Then you could use a solver...

I tried your suggestion @hunse and it sped things up by a factor of ~50x for 50 neurons! I guess cvxpy is not being clever in this case. :( I...

I was not so that's fine by me. You may want to use the above version (without cvxpy) because it's faster, and to also transpose the approach from NnlsL2 to...

For reference sake, a bunch of the above is extended / tested / compared in https://github.com/tcstewar/nengo_solver_dales and featured in https://github.com/astoeckel/nengo-bio.

I like the proposed syntax: ```python nengo.Connection((ens_1x, ens_1y), ens_2, function=lambda x, y: x * y) ``` Nifty! We may also want some notation to factorize the weight matrix and throw...

@tbekolay > Any idea whether these effects would be exaggerated for learning networks? I've often found that learning models (even learning a communication channel) have a hard time accurately learning...