Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Diffusion on syntax trees for program synthesis

tree-diffusion.github.io

Post reply on HN
Print view View original

31–40 of 95 posts

Re: Diffusion on syntax trees for program synthesis

#31

Post by jmugan »

revalo wrote: post #28
jmugan wrote: post #25

[flagged]

It's much easier to work with LISP. It's functional, trees are easy to reason about. We're using an even more simplified language in this work. It's kind of like the MNIST for progsynth research. Scaling this to real and useful programming languages and domains is still non-trivial, and is a major drawback of our work here.

Thanks for your response. I think I would even rather it be a limited subset of Python. But I seem to be in the minority here. Cool work!

Re: Diffusion on syntax trees for program synthesis

#32

Post by gastonmorixe »

could diffusion work at binary level? I mean, could we train a diffusion model to generate a final binary of a program given a prompt? probably AST may be better but the binary I feel is extremely easy to at least test fast if it works or not. Though there may be a lot of drawbacks, if this is possible I can't wait until we ask "give me an app that does that" and the diffusion model starts generating all te bytes the app to do the job. Just wondering

Re: Diffusion on syntax trees for program synthesis

#33

Post by dcreater »

gastonmorixe wrote: post #32

could diffusion work at binary level? I mean, could we train a diffusion model to generate a final binary of a program given a prompt? probably AST may be better but the binary I feel is extremely easy to at least test fast if it works or not. Though there may be a lot of drawbacks, if this is possible I can't wait until we ask "give me an app that does that" and the diffusion model starts generating all te bytes the…

That would be mind blowing. Why go through all the lost intermediary steps, especially through Python and JS, when you can generate machine code directly

Re: Diffusion on syntax trees for program synthesis

#34

Post by treyd »

dcreater wrote: post #33
gastonmorixe wrote: post #32

could diffusion work at binary level? I mean, could we train a diffusion model to generate a final binary of a program given a prompt? probably AST may be better but the binary I feel is extremely easy to at least test fast if it works or not. Though there may be a lot of drawbacks, if this is possible I can't wait until we ask "give me an app that does that" and the diffusion model starts generating all te bytes the…

That would be mind blowing. Why go through all the lost intermediary steps, especially through Python and JS, when you can generate machine code directly

Interpretability, reasoning about the generated code, portability, etc. Probably also demands a larger model with a higher cost of training (and likely more training data) to target a more unstructured "language" like machine code.

Re: Diffusion on syntax trees for program synthesis

#36

Post by eternauta3k »

dcreater wrote: post #33
gastonmorixe wrote: post #32

could diffusion work at binary level? I mean, could we train a diffusion model to generate a final binary of a program given a prompt? probably AST may be better but the binary I feel is extremely easy to at least test fast if it works or not. Though there may be a lot of drawbacks, if this is possible I can't wait until we ask "give me an app that does that" and the diffusion model starts generating all te bytes the…

That would be mind blowing. Why go through all the lost intermediary steps, especially through Python and JS, when you can generate machine code directly

If your model is error-prone, having control structures, types and other compile-time checks is very valuable. It's harder to constrain arbitrary machine code to make something sensible.

Re: Diffusion on syntax trees for program synthesis

#37

Post by zelphirkalt »

This sounds more similar to what people have done with Racket and hint generation for MOOCs. Not sure which university it is again, but I saw a presentation about how they generate hints for students by mutating the syntax tree and analyzing how they had to modify it, to get to a target solution. It was presented at some RacketCon, maybe a decade ago already. Perhaps that knowledge how to do it can be combined with newer machine learning approaches?

EDIT: I found the talk: https://invidious.baczek.me/watch?v=ijyFC36kVis

Re: Diffusion on syntax trees for program synthesis

#38

Post by fulafel »

gastonmorixe wrote: post #32

could diffusion work at binary level? I mean, could we train a diffusion model to generate a final binary of a program given a prompt? probably AST may be better but the binary I feel is extremely easy to at least test fast if it works or not. Though there may be a lot of drawbacks, if this is possible I can't wait until we ask "give me an app that does that" and the diffusion model starts generating all te bytes the…

Editing with feedback from program output, like in this work, could be more closely applicable if you first disassemble the binary and have it edit things in the assembly language AST, then reassemble. This would result in a higher likelihood of creating a valid program.

Re: Diffusion on syntax trees for program synthesis

#39

Post by DougBTX »

adrianmonk wrote: post #9
szvsw wrote: post #8

Earlier quoted context omitted.

> We apply our approach to inverse graphics tasks, where our model learns to convert images into programs that produce those images. I would argue that at least on a philosophical level, this is, definitionally, the process of converting raster graphics to vector graphics, as long as you by the premise that the difference between the two is simply that vector gfx is a programmatic/imperative representation of image g…

Or, raster and vector images are philosophically the same thing . The only difference is that vector has more operations than raster. Raster just has "draw unit square at integer coordinates".

On the other hand, A Pixel Is Not A Little Square[0] would disagree, a raster is a grid sample of a continuous function.

[0] http://alvyray.com/Memos/CG/Microsoft/6_pixel.pdf

Post reply on HN