Shape-constrained symbolic regression

• Grant number: 21/12706-1
• Support Opportunities: Scholarships abroad - Research
• Start date: June 01, 2022
• End date: November 30, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Fabricio Olivetti de França
• Grantee: Fabricio Olivetti de França
• Host Investigator: Gabriel Kronberger
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil
• Institution abroad: University of Applied Sciences Upper Austria, Austria

Abstract

Regression analysis is a statistic tool with the goal of explaining the relationship between measurable variables. This tool can be used for the interpolation and extrapolation of data,identify the properties of the relationship, among other uses. The tools commonly used for this task are either the linear regression, that assumes a linear relationship, or nonlinear models with fixed form like, for example, Neural Network. In some cases, the desired regression model has some required properties. For example, some models are expected to be monotonic w.r.t. one or more of its variables. These constraints may be necessary to ensure that the model obeys observed properties, to ensure equity in treatment, or even improve extrapolation capabilities. To find parametric models that meet these constraints can be challenging since we usually part from a fixed form model and we can only change some adjustable parameters. Symbolic Regression models perform a search in the search space of function forms. Since they do not have a fixed form, this technique allows a greater understanding of the system of interest. Besides that, it is possible to find the function form that contains the desired properties. This research project has the main goal of exploring the shape-constraint symbolic regression algorithms. Until now, this research collaboration has led to 2 international publications, one in the Evolutionary Computation journal and another in NeurIPS conference, both qualis A1. We also have a recently submitted paper under review. (AU)

Scientific publications

(4)

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

HAIDER, C.; DE FRANCA, F. O.; BURLACU, B.; KRONBERGER, G.. Shape-constrained multi-objective genetic programming for symbolic regression. APPLIED SOFT COMPUTING, v. 132, p. 15-pg., 2023-01-01. (21/12706-1)

FERNANDES, MATHEUS CAMPOS; DE FRANCA, FABRICIO OLIVETTI; FRANCESQUINI, EMILIO; PAQUETE, L. HOTGP- Higher-Order Typed Genetic Programming. PROCEEDINGS OF THE 2023 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE, GECCO 2023, v. N/A, p. 9-pg., 2023-01-01. (19/26702-8, 21/06867-2, 21/12706-1)

DE FRANCA, FABRICIO OLIVETTI; BUCHBERGER, B; MARIN, M; NEGRU, V; ZAHARIE, D. Symbolic Regression with augmented dataset using RuleFit. 2022 24TH INTERNATIONAL SYMPOSIUM ON SYMBOLIC AND NUMERIC ALGORITHMS FOR SCIENTIFIC COMPUTING, SYNASC, v. N/A, p. 4-pg., 2022-01-01. (21/12706-1)

DE FRANCA, FABRICIO OLIVETTI; BUCHBERGER, B; MARIN, M; NEGRU, V; ZAHARIE, D. Comparison of OLS and NLS to fit Transformation-Interaction-Rational expressions. 2022 24TH INTERNATIONAL SYMPOSIUM ON SYMBOLIC AND NUMERIC ALGORITHMS FOR SCIENTIFIC COMPUTING, SYNASC, v. N/A, p. 4-pg., 2022-01-01. (21/12706-1)