Symbolic Regression is a powerful technique that seeks to unveil a mathematical expression capturing intricate nonlinear relationships among variables within a given dataset. Its applications extend across diverse scientific domains, including Physics, Astrophysics, Chemistry, Material Science, and more.
Traditionally, Symbolic Regression has been unconstrained in its choice of functional forms to accurately model the data. However, in Scientific Machine Learning the objective is to find an expressive equation that can be understood by the practitioner.
In this project, our objective is to develop a Symbolic Regression algorithm capable of identifying symbolic expressions while adhering to predefined functional forms and shape constraints. These constraints are carefully chosen to align with the specific needs and expectations of the collected data. The endeavor necessitates an interdisciplinary approach, drawing from various scientific disciplines to comprehend the diverse desiderata and constraints at play. We will apply advanced regression analysis concepts to achieve this goal.
Desired knowledge: programming (Haskell, C++, Python), regression analysis
- A greedy search tree heuristic for symbolic regression
- Interaction–transformation evolutionary algorithm for symbolic regression
- Transformation-interaction-rational representation for symbolic regression
- Transformation-Interaction-Rational Representation for Symbolic Regression: A Detailed Analysis of SRBench Results