Closed studywolf closed 9 years ago
Hmm... I think I'd like to argue that the benchmarking specification part of this paper should be agnostic to how you program the hardware. People should be able to use this benchmark regardless of how they do the programming. Indeed, as long as their hardware can talk to Python, it should be fine to run. I'm vaguely tempted to restructure the benchmark_min code a bit to show explicitly how to do that, since that'd be straightforward, but it's a bit hard to structure when I don't know what a non-nengo generic interface would be... hmm....
That said, when I do the analysis section, I'll definitely say that it was easy to implement this because Nengo and talk about it a bit.
right! sorry after reading through how the task is described below and the use of the hardware for the basis function generation i withdraw this comment! it's probably just the angle i'm coming from, maybe a mention that all that's needed from the hardware is the activation in response to an input signal would be good, but maybe that's also just assumed by everyone already!
https://github.com/tcstewar/2015-Embodied_Benchmarks/blob/master/paper/paper.tex#L261
suggest referencing here something, maybe in a footnote, about software like Nengo that is available to use to program this hardware, as that would be a question i might have if i was developing neuromorphic hardware. 'i'm just doing the developing of the hardware, how can i program it to run these benchmarks without taking another 3 years to program it' or something