Symbolic Regression and Polynomial Optimization in Automated Scientific Discovery

Abstract

Recently, there have been lots of work done in designing both symbolic regression and polynomial optimization tools for discovering equations in physics both with and without background knowledge. Given some physical data {(x_1,…,x_n,y)_i}, how can we learn a “nice” function f(x_1,…,x_n) with some background knowledge B? In this talk, I’ll explore some recent work in 1) querying neural networks or a black box model to learn a function piece by piece and 2) setting up scientific discovery as a polynomial optimization / LP problem, with a cameo from algebraic geometry. Time permitting, I’ll talk a bit about some ongoing work in this space.

Karan Srivastava

PhD Student, Mathematics

My research interests include machine learning, reinforcement learning, combinatorics, and algebraic geometry