Deep Learning for Symbolic Mathematics. Guillaume Lample, François Charton

48 views
Skip to first unread message

alex.shkotin

unread,
Dec 9, 2019, 4:19:03 AM12/9/19
to ontolog-forum
FYI
This paper about ANN trained to integrate and differentiate symbolically is absolutely technical. But their conclusion on p. 12 awakes the idea that maybe they have a model of intuition:-)
"We show that a simple transformer model
trained on these datasets can perform extremely well both at computing function integrals, and
solving differential equations, outperforming state-of-the-art mathematical frameworks like Matlab or
Mathematica that rely on a large number of algorithms and heuristics, and a complex implementation
(Risch, 1970)." 

Alex
Reply all
Reply to author
Forward
0 new messages