Edge Rewrite
// HTMLRewriter · presentation

This page was redesigned at the edge.

Cloudflare fetched the original article and streamed it through HTMLRewriter to apply an entirely new visual system without rebuilding the source page.

Jump to content

Talk:Differentiable programming

Page contents not supported in other languages.
Add topic
From Wikipedia, the free encyclopedia

the abbreviation “∂P”

[edit]

An anonymous user reverted my change removing the abbreviation “∂P” from this page. To be clear, I don’t think this abbreviation meets Wikipedia’s standards for reliable sources: https://en.m.wikipedia.org/wiki/Wikipedia:Reliable_sources. There is exactly one source that uses this abbreviation, and it’s an unpublished and unreviewed paper that only appears on the arXiv. S. Hoyer (talk) 08:11, 4 November 2019 (UTC)Reply

I completely agree with you and reverted the reversion :-) To be clear, I think it's reasonable that the “∂P” paper used the abbreviation, yet it does not seem to be a standard and so there's no justification for putting it throughout the article. Sanpitch (talk) 23:48, 5 November 2019 (UTC)Reply

What *is* differentiable programming, exactly?

[edit]

I‘m a CS major, but after reading the article, I have NO clue what the lemma is about. I can‘t even begin to imagine what differentiation could mean in a programming language context. This article could use some work... Crazor (talk) 21:35, 13 April 2020 (UTC)Reply

Dittoes. It's unclear at first glance whether "programming" refers to computer programming, or to linear programming. I suppose when it says "differentiable," it's talking about automatic differentiation, but somebody popped in a link to the calculus concept of "Derivative," just to muddy the water. 173.75.33.51 (talk) 21:31, 8 July 2020 (UTC)Reply
Yes. I happened on the article serendipitously. And I leave no wiser than when I came. CS background too. Some clue as to both the meaning and the benefits or purpose of differentiation in this context is required. 5.181.56.1 (talk) 13:09, 12 January 2023 (UTC)Reply
Here's what I think it is or could be - machine learning where backpropagation-like training doesn't just optimize weights/parameter of fixed functions, but also figures out which functions to use (among a specified set) and in whatever sequence. The "which" question can be solved with gating functions. However, "sequence" - I have no idea. Msltoe (talk) 01:20, 19 September 2023 (UTC)Reply

Static vs dynamic distinction

[edit]

This distinction seems a bit arbitrary and e.g. Myia doesn't use it, it is just source-to-source vs operator overloading. IMO the distinction should be removed, just discuss the approaches chronologically. Thoughts? Mathnerd314159 (talk) 16:31, 19 July 2022 (UTC)Reply