←back to thread

Backpropagation is a leaky abstraction (2016)

(karpathy.medium.com)

346 points swatson741 | 1 comments | 02 Nov 25 05:20 UTC | HN request time: 0.219s | source

Show context

drivebyhooting ◴[02 Nov 25 07:21 UTC] No.45788473[source]▶

>>45787993 (OP) #

I have a naive question about backprop and optimizers.

I understand how SGD is just taking a step proportional to the gradient and how backprop computes the partial derivative of the loss function with respect to each model weight.

But with more advanced optimizers the gradient is not really used directly. It gets per weight normalization, fudged with momentum, clipped, etc.

So really, how important is computing the exact gradient using calculus, vs just knowing the general direction to step? Would that be cheaper to calculate than full derivatives?

replies(11): >>45788484 #>>45788525 #>>45788559 #>>45788604 #>>45789234 #>>45789345 #>>45789485 #>>45792574 #>>45792998 #>>45796071 #>>45799773 #

1. killerstorm ◴[03 Nov 25 15:02 UTC] No.45799773[source]▶

I don't think there's a way to find "general direction" without actually using calculus.

If you write down an explicit expression for partial derivatives, it will contain sums. The sign (which kind of defines the general direction) is affected by what's in the sum, and you can't avoid calculating it.