Multi-Task Growing Interpretable Neural Network For Multi-Target Symbolic Regression

Author

• Hussein Rajabu, Prairie View A&M University

Date of Award

5-2026

Document Type

Thesis

Degree Name

Master of Science

Department

Electrical Engineering

Abstract

Over the past decade, deep learning has achieved remarkable success across domains such as computer vision and natural language processing. Despite their strong predictive capabilities, these models often function as black boxes, limiting transparency and making them difficult to trust in high-stakes applications such as healthcare and business decision-making. While explainable artificial intelligence (XAI) methods improve interpretability, most approaches remain post-hoc, focusing on statistical associations rather than uncovering the underlying functional mechanisms governing predictions.

Symbolic regression (SR) has emerged as a promising direction within XAI by directly discovering explicit mathematical expressions that describe relationships between inputs and outputs. Unlike traditional black-box models, SR provides interpretable equations, enabling both accurate prediction and mechanistic understanding. However, existing SR approaches face two major limitations: they are primarily validated on scientific datasets with well-defined relationships, and they largely focus on single-target regression, limiting their applicability to real-world problems involving multiple interdependent outputs

To address these challenges, this thesis proposed Multi-Task Regression GINN-LP (MTRGINN-LP), a neuro-symbolic framework for multi-target symbolic regression. The model extended the Growing Interpretable Neural Network with Laurent Polynomial structure (GINN-LP) by integrating multi-task learning. It employed a shared backbone of Power-Term Approximator Blocks (PABs), which captured multiplicative and power-law relationships, followed by task-specific linear output layers. This design enabled the discovery of shared symbolic structures across targets while preserving task-specific interpretability. Additionally, a symbolic consistency loss is introduced to align neural predictions with their corresponding symbolic representations during training.

The proposed method is evaluated on practical multi-target tasks, including energy efficiency prediction and sustainable agriculture. Experimental results demonstrate that MTRGINN-LP achieves competitive predictive performance while maintaining strong interpretability, effectively bridging the gap between symbolic regression and real-world multi-output learning scenarios.

Index Terms—Interpretable artificial intelligence, Laurent polynomial, multi-task learning, neuro-symbolic artificial intelligence (AI), symbolic regression.

Committee Chair/Advisor

Xishuang Dong

Committee Member

Lijun Qian

Committee Member

Xiangfang Li

Publisher

Prairie View A&M University

Rights

© 2021 Prairie View A & M University

Date of Digitization

7/2/2026

Contributing Institution

John B Coleman Library

City of Publication

Prairie View

MIME Type

Application/PDF

Recommended Citation

Rajabu, H. (2026). Multi-Task Growing Interpretable Neural Network For Multi-Target Symbolic Regression. Retrieved from https://digitalcommons.pvamu.edu/pvamu-theses/1678