Reproduce “Bridging Data Gaps in Structural Fragility Modeling through Transfer Learning: Methodology and Case Studies” on OpenResearch

About this paper

Narges Saeednejad, Jamie Ellen Padgett

This research introduces a transfer learning framework to adapt structural fragility models for data-scarce settings, improving predictive performance and reliability. The framework, applied through four distinct strategies, reduces bias and enhances failure detection in low-data regimes while maintaining engineering interpretability and quantifying uncertainty.

Key insights

• Transfer learning can effectively bridge data gaps and mitigate domain shift in structural fragility modeling, substantially outperforming direct model transfer.
• Different transfer learning strategies (e.g., instance-based reweighting, parameter fine-tuning, hierarchical Bayesian priors, multi-source fusion) are best suited for specific data scarcity and domain shift scenarios.
• Incorporating local covariates and leveraging existing knowledge from multiple source models significantly improves the accuracy and stability of fragility predictions.

Summary by alphaXiv.