Deep Learning isn’t the brain
Sean Aubin
405

I personally think it’s only a matter of time before a biologically plausible method is discovered

Forward-prop?

If there’s a lot of neurotransmitter still floating around the synapse, perhaps that results in a sort of weighted ‘forward’-prop that drags performance down the correct path, whilst the lack of it feeds back, so to speak, as a relatively negative ‘back’-prop — it’s not that back-prop happens as such but the forwardly weighted neurons fire sooner and, so, the less ‘charged’ connections effectively lose out in the competition to influence the outcome. [1]

The less influential inputs become even less positively weighted … resulting in an effective back-prop mechanism, albeit that it isn’t strictly back-prop as we know it in Comp.Sci. terms.

I’m being a bit vague … just musing really … but I could see it working that way.

[1] “Thanks for your input, but we’ve already found the solution.”

Show your support

Clapping shows how much you appreciated Where Angels Fear’s story.