Papers of the day   All papers



Sep 26 2019 Ilya Sutskever

Really enjoyed the (non OpenAI) ICLR submission that trained a transformer on symbolic math. The surprise: it beat Mathematica on symbolic integration and diff eq solving by a _very_ big margin!
12 replies, 1094 likes

Sep 26 2019 Erik Bernhardsson

This paper uses seq2seq to do symbolic integrals by training it on (f’(x), f(x)) pairs and it’s such an obviously great idea that I wish I thought of first! I can’t be the only one having that feeling.
10 replies, 157 likes

Sep 27 2019 Christian Szegedy

Approximate mathematical reasoning is possible in the latent space alone. We created semantic embedding of formulas and performed complicated multi-step reasoning on them, then we compared it with the symbolic results:
3 replies, 154 likes

Sep 27 2019 algoritmic

Deep Learning for Symbolic Mathematics #math
2 replies, 115 likes

Sep 26 2019 Hacker News

Deep Learning for Symbolic Mathematics
0 replies, 38 likes

Nov 03 2019 Stephen Diehl

This is one of the wilder applications I've seen for deep learning. Page 8: "This suggest that some deeper understanding of [differential equation] mathematics has been achieved by the model."
1 replies, 25 likes

Sep 27 2019 P. Oscar Boykin

What I like about this is that it feels like how people solve inverse problems: use ✨intuition✨ to guess a solution, then check that it is correct (and maybe apply a delta to solve rest). Seems like this paper show how to implement a computer algebra system to do that.
0 replies, 17 likes

Sep 26 2019 Timothy Gowers

Trying to decide where this appears on the spectrum that runs from quite amusing at one end to game changer at the other. At a first glance it seems pretty impressive.
3 replies, 15 likes

Sep 26 2019 Xander Steenbrugge

This is brilliant & absolutely stunning! After all these years of Deep Learning I'm still amazed at what a neural net can do when you provide the right input representations & throw stupendous amounts of computation at it... 😲👏
0 replies, 14 likes

Oct 02 2019 Cam DP

This but for deconvolution problems, too.
0 replies, 6 likes

Sep 26 2019 Eclipse DL4J

0 replies, 4 likes

Sep 27 2019 Barney Pell

Wow! Deep learning performing much better than symbolic math solvers.
0 replies, 3 likes

Sep 26 2019 Peter Bloem

With just 6 layers. You could probably fit the whole model in 12Gb of GPU memory.
0 replies, 3 likes

Sep 26 2019 Melanie Mitchell

0 replies, 1 likes

Sep 26 2019 mats-erik pistol, the man in the middle.

For a while I thought it was April 1st. Transformer networks beat Mathematica by a large margin when doing symbolic integration and finding solutions to differential equations. The singularity is near:
1 replies, 0 likes