2
$\begingroup$

I am attempting to solve quadratic equations using machine learning. I tried to write code using mxnet but fails because it is using linear regression while my problem is non-linear (as far as I know).

It is possible to predict the solution for arbitrary equations provided the at least one real root exists?

$\endgroup$
3
  • $\begingroup$ Could you give an example of the form that you receive the equation in? For instance are you receiving (x, y) data and want to fit a quadratic equation to it? Or are you receiving an equation $ax^2 + bx + c = 0$? $\endgroup$ Jan 24, 2018 at 14:18
  • $\begingroup$ @NeilSlater The equation itself (i.e. your 2nd option). This is just for testing machine learning capablitity for predicting solutinos for such problem where the solution is unknown. $\endgroup$ Jan 25, 2018 at 10:02
  • $\begingroup$ This should answer your question at least in part: researchgate.net/publication/… $\endgroup$
    – Vitaliy
    Mar 18, 2019 at 18:38

1 Answer 1

2
$\begingroup$

It is possible to predict the solution for arbitrary equations provided the at least one real root exists?

Yes.

You‘ll just need a sufficiently large training dataset and network that is flexible enough (start with one hidden layer).

You’re right your problem is not linear, thus you’ll need non-linear activation function.

$\endgroup$
1
  • $\begingroup$ Thanks, but I tried many times to build a fully connected network without success getting even 0% accuracy on +100k dataset built using counter that evaluates in a quadratic equation. $\endgroup$ Apr 4, 2019 at 10:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.