Robust Strength Behavior Modeling of Coarse-Grained Soils Using HSIC-Guided Stable Learning

ICLR 2026 Conference Submission9391 Authors

Keywords: Coarse-grained soils; Strength behavior; Hilbert-Schmidt Independence Criterion; Stable learning ; Out-of-distribution

Abstract: Coarse-grained soils are widely employed in infrastructure construction, and capturing their strength behavior is vital for ensuring the structural integrity of engineering systems. In recent years, artificial intelligence (AI) techniques have shown significant promise in advancing investigations in this area. Nevertheless, conventional AI models often exhibit limited robustness when confronted with distributional shifts in the data. To tackle these limitations, this study introduces a stable learning framework based on the Hilbert-Schmidt Independence Criterion, referred to as HSIC-StableNet, for predicting deviatoric stress–axial strain curves that represent the strength characteristics of coarse-grained soils. The proposed method initially adopts HSIC with the exact kernel method to replace the F-norm combined with the approximate kernel method, strategically reweighting training samples to enhance the stable learning module and integrating it with a deep neural network. The experimental results indicate that HSIC-StableNet consistently surpasses conventional DNN models and a previously introduced stable learning approach, SNN, across key metrics such as R², MSE, MAE, and MAPE. Furthermore, the model demonstrates strong performance in estimating the strength behavior of coarse-grained soils with large particle sizes by utilizing data samples from soils with smaller particles. This capability contributes to alleviating the data scarcity challenge in geotechnical engineering, where acquiring adequate large-particle soil data through costly triaxial tests remains difficult.

Primary Area: causal reasoning

Submission Number: 9391