Re:Show HN: Deep Learning for Program Synthesis
by suryabhupa on Friday April 21, 2017 @02:49PM
(#14168221)
One of the authors here -- would love to answer any questions about the work! :)
[ Parent
| on HN ]
Re:Show HN: Deep Learning for Program Synthesis
by lostmsu on Friday April 21, 2017 @08:01PM
(#14170380)
Are you going to publish the source code for reproduction?
Re:Show HN: Deep Learning for Program Synthesis
by suryabhupa on Saturday April 22, 2017 @12:29PM
(#14173258)
Eventually, yes.
Re:Show HN: Deep Learning for Program Synthesis
by harigov on Friday April 21, 2017 @07:32PM
(#14170258)
Are there any difficulties in generating a program in standard languages in Python? Did you choose a DSL because neural network is sensitive to the output programming language?
Re:Show HN: Deep Learning for Program Synthesis
by suryabhupa on Saturday April 22, 2017 @12:31PM
(#14173260)
It turns out the full grammar of Python (and almost all real programming languages) is quite large; this is very early and new work in neural program synthesis, and so we chose a pretty limited DSL to make sure that we could at least solve this one before moving on to more general ones that contain state, conditionals, for-loops, etc. In theory however, we can apply the exact architecture to Python programs and see what happens. We haven't tried yet. :)
Re:Show HN: Deep Learning for Program Synthesis
by henning on Friday April 21, 2017 @06:20PM
(#14169882)
The first thing I thought of when reading this article was genetic programming.
Is this a significant improvement over evolutionary computation methods? Has that been attempted in the past?
Re:Show HN: Deep Learning for Program Synthesis
by suryabhupa on Saturday April 22, 2017 @12:35PM
(#14173284)
I'm not too familiar with evolutionary computation methods, but I imagine the approaches may be similar in nature.
Re:Show HN: Deep Learning for Program Synthesis
by Gargoyle on Friday April 21, 2017 @10:48PM
(#14170873)
Why does the final example in figure 14 fail completely? The outputs are correct as far as they go, but they're all incomplete.
Is it because the scoring metric has a point where enough of a good start outscores an alternative in the beam search that could lead to a more complete solution? In non-trivial real-world examples, would the be a major problem?
Re:Show HN: Deep Learning for Program Synthesis
by krapht on Friday April 21, 2017 @08:33PM
(#14170494)
Any chance this could be implemented for automated proof synthesis? Would love to see this used in F* or Lean.
Re:Show HN: Deep Learning for Program Synthesis
by suryabhupa on Saturday April 22, 2017 @12:33PM
(#14173273)
Theorem solving is very closely related to program induction (we just change the grammar). Just as with Python, the underlying search space would be incredibly large, and while in theory, we could simply change the DSL and it should work, it'll probably involve a few more iterations of the model or other insights to see this to fruition (but it's definitely not impossible).
Re:Show HN: Deep Learning for Program Synthesis
by kyberias on Saturday April 22, 2017 @05:52AM
(#14171788)
What is the link with Excel FlashFill feature?
Re:Show HN: Deep Learning for Program Synthesis
by mldeeplearn on Sunday April 23, 2017 @05:40AM
(#14177370)
It looks like this is a more comprehensive version of FlashFill (can do more tasks), and it is based on deep learning instead of previous rule-based techniques in FlashFill.