Bayesian Filter Proposed for Noisy Geotechnical Data

Researchers propose adapting labeled random finite sets, a concept from multi-object tracking, to manage the inherent uncertainty in geotechnical databases, potentially strengthening data-driven infrastructure analysis.

By Adrian Cole

Chips & Infrastructure

October 19, 2026 at 12:03 PM EDT·1 min read

Key Takeaways (3)

•The method applies the GLMB filter—a Bayesian multi-object estimator—to geotechnical property estimation for the first time.
•The preprint tests the approach on a public clay database but does not disclose quantitative performance results.
•The paper is an unreviewed preprint, and independent validation has not been established.

Key Takeaways

•The method applies the GLMB filter—a Bayesian multi-object estimator—to geotechnical property estimation for the first time.
•The preprint tests the approach on a public clay database but does not disclose quantitative performance results.
•The paper is an unreviewed preprint, and independent validation has not been established.

A preprint posted to arXiv on October 19 introduces a method for estimating soil and rock properties from uncertain geotechnical databases using labeled random finite sets (LRFS).

The work addresses a long-standing practical challenge: geotechnical datasets—subsurface measurements of soil strength, compressibility, and other parameters—are often noisy and incomplete. That uncertainty can propagate into the data-driven models that engineers increasingly use to assess foundation risk, slope stability, and tunnelling conditions, making infrastructure decisions less reliable.

The paper proposes framing geotechnical observations as objects within an LRFS framework and applying the generalized labeled multi-Bernoulli (GLMB) filter, a Bayesian multi-object estimator originally developed for target tracking. According to the abstract, the similarity between LRFS characteristics and the “big indirect data” (BID) typical of geotechnical databases allows the filter to handle measurement uncertainties probabilistically.

Numerical experiments are performed on a public clay database to evaluate the approach. The abstract does not report specific accuracy metrics, benchmark comparisons, or independent replication. The paper is a preprint and has not been peer-reviewed.

Bayesian Filter Proposed for Noisy Geotechnical Data

Key Takeaways

More from Automatica Press

SAGE-CP preprint proposes route-stratified calibration for adaptive VLM ensembles

AWS Trainium Adds Native PyTorch Backend, Hugging Face Integration

Preprint proposes In-Context Learning beyond pretraining support