Comment by StandardFuture | Hacker News Reader

Comment by StandardFuture | Hacker News Reader

Parent Full thread

> any thread unifying these things

Yes, this is the critical question. And I think the authors of this "reading group" answer that in this really smooth introduction to sequent calculus:

https://www.youtube.com/watch?v=xLRUofFSq5Y

Recommended watching!

The summary (explained in the video) is that these three areas: differential programming, probabilistic programming, and logic programming are all just programming language semantics that are more specifically "languages for calculating" and thus "calculi" languages.

These three areas are also all linked to historical and current AI research. And I think AI research is the primary motivation (another underlying thread) for this collection.

Thank you for your kind remarks! As you mentioned, a better name might have been "languages for calculation". In French these subjects are all "calcul". Personally, I would skip the first few and listen to the guest speakers, who have spent a lot more time thinking about these topics. You might also enjoy the lectures from OPLSS, where I learned about many of these ideas. [1]

  These three areas are also all linked to historical and current AI research.

Indeed, they all seem to have some fascinating connections to symbolic and connectionist learning. I think the symbolists were originally on the right track, but they were too preoccupied with expert systems when they should have been working at a much lower level of representation. One way to view the current wave of deep learning is as a form of symbolic differentiation with a special order of evaluation corresponding to reverse-mode AD.

[1]: https://www.youtube.com/channel/UCDe6N9R7U-RYWA57wzJQ2SQ

Reply on news.ycombinator.com