Gradients without Backpropagation · HackerPapers discussions

Gradients without Backpropagation — mentioned by arugulum

Wed, 07 Jun 2023 02:20:53 GMT

Right, I hate to be that guy, but it essentially boils down to "take a random step and see if it's good, if so update" (with some fanciness). It should go without saying that this is horribly inefficient in a very high dimensional space (e.g. the number of parameters). The results are decent, but it's just another in a long line of work (and more recent work https://arxiv.org/abs/2202.08587 https://openreview.net/pdf?id=5i7lJLuhTm) trying to use backprop algorithm with some promising results but still falling far short of the standard backprop workhorse.

Paper and discussion history. A mention is not an endorsement.

Gradients without Backpropagation — HN submission

Wed, 02 Mar 2022 08:41:05 GMT

Gradients Without Backpropagation

87 points; 35 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Gradients without Backpropagation — HN submission

Sun, 20 Feb 2022 16:36:54 GMT

Gradients Without Backpropagation

4 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Gradients without Backpropagation — HN submission

Fri, 18 Feb 2022 03:12:48 GMT

Gradients Without Backpropagation

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.