Re: [Maxima-discuss] ML for symbolic math

[Richard F. <fa...@be...>]

I looked at this paper, but did not study it closely.
   It makes substantial
use of symbolic math (sympy) to simplify the random expressions
it generates.  Given such an expression S, it then
differentiates it, to get dS.   It then essentially
memorizes that the integral of dS is S.
The test set offered to this "ML" program, to Mathematica, and to Matlab
is to integrate the examples dS1, dS2, .....  (at least it looks that 
way to me...)

On this rather artificial set, the ML program does much better than
the CAS programs.  My guess is that for one of the integration problem sets
used to compare CAS capabilities, this ML program would get about 0%.

Another task, solving ODEs, is also posed.

Given that this example of ML requires a simplifier, and the test set is
100% biased, I think this is not going to be of interest in the symbolic 
math
system building community.

Has anyone else looked at this paper?.

RJF

On 9/29/19 5:58 PM, Elias Mårtenson wrote:
> Here's a paper titled “Deep learning for symbolic mathematics” that 
> might be of interest to the members of this mailing list.
>
> https://openreview.net/pdf?id=S1eZYeHFDS
>
> Regards,
> Elias
>
>
> _______________________________________________
> Maxima-discuss mailing list
> Max...@li...
> https://lists.sourceforge.net/lists/listinfo/maxima-discuss