Enhancing seismic risk assessment: a sparse multi-task learning approach for high-dimensional structural systems

Accurate quantification of engineering demand parameters is essential for probabilistic risk assessment and performance evaluation of structural systems, particularly for multi-response structures subjected to seismic events.Complex real-world structures often exhibit multiple demand parameters corresponding to distinct failure modes, necessitating a robust modeling approach that captures these interdependencies effectively.Conventional univariate models, which predict individual seismic demand responses, fail to account for the inherent correlations among multiple structural responses, leading to suboptimal predictions.Alternatively, multivariate modeling frameworks leveraging multi-task learning (MTL) techniques have emerged as a promising solution, enabling the simultaneous and joint prediction of multiple responses by improving generalization performance through the sharing of information across tasks.Despite advancements in this area, effectively capturing intricate task correlations, particularly within high-dimensional feature spaces, and mitigating the influence of noise covariance remain significant challenges.To address these challenges, this study proposes an innovative MLT framework enhanced with sparse network Lasso regularization.The proposed approach integrates MTL principles with a sparsity-inducing regularization technique through a novel penalty function that encourages highly correlated tasks to exhibit similar sparsity patterns.This integration strategically enforces sparsity in the coefficient matrix, effectively reducing the impact of irrelevant or noisy features and enhancing the resilience of the model to outliers.By leveraging this sparsity-driven regularization, the framework improves task correlation modeling and delivers precise multi-response seismic demand predictions, even in complex, high-dimensional feature spaces.As a result, the proposed framework notably advances the precision and reliability of probabilistic seismic demand analysis for multi-response structures such as bridges.By providing robust and accurate estimates of critical engineering demand parameters, it supports not only the design of new structures but also the evaluation, retrofitting, and maintenance of existing structures.This novel approach plays a crucial role across all phases of seismic hazard management, preparedness, response, repair, and recovery, by enabling informed decision-making that strengthens infrastructure resilience.Ultimately, the framework contributes to enhancing the Sonia Zehsaz, Farahnaz Soleimani 2 robustness of bridge systems against seismic hazards, safeguarding community safety, and promoting long-term infrastructure sustainability in the face of increasing natural hazard risks.