![[2602.20194] FedAvg-Based CTMC Hazard Model for Federated Bridge Deterioration Assessment](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20194] FedAvg-Based CTMC Hazard Model for Federated Bridge Deterioration Assessment
Summary
This article presents a federated framework using a CTMC hazard model for assessing bridge deterioration, allowing municipalities to collaboratively train models without sharing sensitive data.
Why It Matters
The proposed model addresses the challenge of data governance in public infrastructure by enabling collaborative analysis while preserving data privacy. This innovation can enhance infrastructure maintenance strategies and improve safety without compromising sensitive information.
Key Takeaways
- Federated learning allows municipalities to train a shared model without sharing sensitive inspection data.
- The CTMC hazard model focuses on key deterioration transitions, improving predictive accuracy.
- Local optimization techniques enhance model training efficiency and convergence.
- The framework incentivizes data sharing by providing users with updated global benchmark parameters.
- Simulation results demonstrate effective convergence across heterogeneous user data.
Computer Science > Machine Learning arXiv:2602.20194 (cs) [Submitted on 22 Feb 2026] Title:FedAvg-Based CTMC Hazard Model for Federated Bridge Deterioration Assessment Authors:Takato Yasuno View a PDF of the paper titled FedAvg-Based CTMC Hazard Model for Federated Bridge Deterioration Assessment, by Takato Yasuno View PDF HTML (experimental) Abstract:Bridge periodic inspection records contain sensitive information about public infrastructure, making cross-organizational data sharing impractical under existing data governance constraints. We propose a federated framework for estimating a Continuous-Time Markov Chain (CTMC) hazard model of bridge deterioration, enabling municipalities to collaboratively train a shared benchmark model without transferring raw inspection records. Each User holds local inspection data and trains a log-linear hazard model over three deterioration-direction transitions -- Good$\to$Minor, Good$\to$Severe, and Minor$\to$Severe -- with covariates for bridge age, coastline distance, and deck area. Local optimization is performed via mini-batch stochastic gradient descent on the CTMC log-likelihood, and only a 12-dimensional pseudo-gradient vector is uploaded to a central server per communication round. The server aggregates User updates using sample-weighted Federated Averaging (FedAvg) with momentum and gradient clipping. All experiments in this paper are conducted on fully synthetic data generated from a known ground-truth parameter set with regio...