The Simple Essence of Automatic Differentiation - Conal Elliott

Channel:Microsoft Research · Length:1:30:45 · Published:2018-07-26 · Views:17062

Watch on YouTube

Summary

Conal Elliott presents his paper on automatic differentiation, deriving a simple, generalized AD algorithm from a natural specification. Varying the representation of derivatives yields forward and reverse mode algorithms that avoid the complexity and statefulness of backpropagation.

Why HN recommends it

Cited as the more accessible companion to the paper, and as the most interesting treatment of AD via compiling to categories.

There's a talk that Conal gives that makes the paper a bit more accessible — HN, 2022

Functional programming · Machine learning & deep learning · Pure mathematics

Subtitles (excerpt)

>> I'm Daan and I'm honored to host today Conal Elliott, former MSR colleague and a good friend of mine. Conal used to work here at Microsoft Research until 2002, and now he's a distinguished scientist at the target. But the way he was here at Microsoft research, I was so busy students and I remember being so inspired by his beautiful work on functional reactive animations, and at that time was really inspirational to me. One thing that Conal is great at is taking something really complicated. In that time like animations you will always have matrixes and frames and timers, and he kind of rephrase it in a beautiful way about the essence of animations as functions of time. I think with this work he applies the same trick again but now in terms of differentiation and machine learning, it's always about all these complicated things, and you almost don't see the essence, and this talk will hopefully shed some light on it. So, thanks a lot, Conal. >> Thanks, Don. It's great to be here. Yes, I was at Microsoft Research in the graphics group from '94 to 2002, I had a good time. So, this is my first time back giving a talk. So, anyway, we're going to talk about automatic differentiation which has

On Hacker News

The most interesting thing I've seen on AD is "The simple essence of automatic differentiation" (2018) [1]. See past discussion [2], and talk [3]. I think the main idea is that by compiling to categories and pairing up a function with its derivative, the pair becomes trivially composable in forward mode, and the whole structure is easily converted to reverse mode afterwards. [1]: [LINK] [2]: [LIN… — infogulch, 2022-04
I'm yet to give reverse mode a second try, but my attempts at forward mode (linked in parent post) were on my own without the paper's help. There's a talk that Conal gives that makes the paper a bit more accessible - [VIDEO] - but only a bit more. I know it'll take effort to wrap my head around the implementation details ... but the paper certainly helped clarify things I was stumbling around abo… — sriku, 2022-09
He also gave nice presentation on AD: [VIDEO] (surprised that you didn't already mention it, considering the venue ;-)) — noblethrasher, 2019-11
Another paper related to AD was posted last month: The Simple Essence of Automatic Differentiation [0]. Video presentation [1]. I found it pretty awesome, just in case you hadn't seen it. [0]: https//news.ycombinator.com/item?id=18306860 [1]: [VIDEO] — infogulch, 2018-11
This paper won a best paper award at ICFP 2018. Conal Elliott recently gave a longer talk about it at Microsoft Research: [VIDEO] — bmc7505, 2018-10

Related videos