Steering Neural Network Training through Interpretable Constraints Based on Partial Dependence

2026-07-09

AI Summary

This work proposes a novel approach to incorporate domain knowledge into neural network training by enforcing user-specified functional priors as partial dependence constraints, thereby aligning the model’s average response to specific features with prior expectations. For the first time, partial dependence analysis is reformulated as a differentiable training constraint, enabling joint optimization of model interpretability and predictive performance. Experimental results across multiple regression tasks—including dynamical system forecasting—demonstrate that the proposed constrained models not only achieve substantially improved prediction accuracy and data efficiency but also ensure that model explanations remain consistent with established domain knowledge.

Abstract

Over the last few years, there has been an increased interest in making machine learning models more interpretable. Although a great deal of effort goes into developing techniques for interpreting the interactions learned by a given model, fewer studies focus on assessing the quality of such explanations. Even fewer focus on how to adjust the model to produce explanations faithful to prior knowledge, a process known as explanation-guided learning. Furthermore, most approaches in this area focus on classification problems and usually assume prior knowledge about which input features or regions are most important. In this work, we introduce a new approach to steering neural networks based on partial dependence, such that their average response to certain features aligns with specific functional domain knowledge about the problem. We empirically demonstrate on a range of regression problems, including dynamical systems forecasting, that models whose training has been controlled using our method perform better than unconstrained models and are more data-efficient. Moreover, we highlight that interpretations obtained from the former actually align with the user-provided knowledge, whereas those obtained from the latter do not.

Problem

• interpretability
• explanation-guided learning
• partial dependence
• domain knowledge
• neural network training

Innovation

• explanation-guided learning
• partial dependence
• interpretable constraints
• neural network steering
• domain knowledge alignment