eh, I'm not really convinced that there would be no observable correlation between the direction the error is moving in, and the direction parameters are moving in. That's pretty much exactly what SGD does.
Either way, I certainly hope you can see how that single sentence you put there is not going to make anyone at all think of temporal effects in activation levels of neurons.
And yeah, that's the other thing; you say "attenuation matrix" is the correct term, I'd still love to see even a single source of someone in AI using that name for a weight matrix. Also, above you used "attenuation…