Prior-Guided Symbolic Regression: Towards Scientific Consistency in Equation Discovery

Jing Xiao, Xinhai Chen, Jiaming Peng, Qinglin Wang, Menghan Jia, Zhiquan Lai, Guangping Yu, Dongsheng Li, Tiejun Li, Jie Liu · 2026

Read Prior-Guided Symbolic Regression: Towards Scientific Consistency in Equation Discovery by Jing Xiao, Xinhai Chen, Jiaming Peng, Qinglin Wang, Menghan Jia, Zhiquan Lai, Guangping Yu, Dongsheng Li, Tiejun Li, Jie Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-13021.txt.

Description

Symbol regression hunts for formulas behind data, but discovered formulas often fit well while violating basic science, a trap the authors call pseudo-formulas. Jing Xiao and nine co-authors propose PG-SR, a prior-guided system that bakes domain knowledge into discovery. A three-stage pipeline of warm-up, evolution, and refinement carries a prior constraint checker, coding domain priors as executable programs, plus a prior annealing testing workings that steers evolution toward scientifically steady regions.

Theory shows PG-SR shrinks the Rademacher complexity of the guess space, tightening generalisation bounds and guaranteeing protection against pseudo-formulas. Tests across diverse domains show PG-SR beating state-of-the-art plain rivals while staying robust to imperfect priors, noisy data, and scarce data. For formula discovery, the message is that priors, properly enforced, improve both trust and closeness.

Catalog record

CREATORJing Xiao, Xinhai Chen, Jiaming Peng, Qinglin Wang, Menghan Jia, Zhiquan Lai, Guangping Yu, Dongsheng Li, Tiejun Li, Jie Liu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-13021

Browse the full library · About the Tower of Babel