Learning Earthquake Wave Arrival Time Picking from Labels with Inaccuracies

2026-06-13 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI in Geophysics · Depth: Expert, quick

Summary

A new Label Noise-Contrastive Robust Learning (LaNCoR) approach has been developed to mitigate the impact of inaccurately labeled training data, or "label noise," in supervised machine learning models. This method is particularly relevant for seismic signal processing tasks, where traditional solutions like large-scale training sets or data augmentation are labor-intensive and expensive. LaNCoR operates by aligning the input waveform feature and label representation distributions within the feature space, effectively correcting mislabeling and reducing its detrimental effects during the training process. Evaluated on the task of P-phase arrival-time picking using real microseismic data and two baseline models, LaNCoR demonstrated significant performance improvements, achieving up to 28.8% better results across various metrics. This approach offers a promising alternative for robust model training in seismology and geosciences.

Key takeaway

For Machine Learning Engineers developing models for seismological applications, especially when dealing with noisy labeled data, you should consider integrating the LaNCoR approach. This method offers a robust alternative to costly large-scale datasets or data augmentation, directly improving model generalization and accuracy. Implementing LaNCoR can significantly enhance P-phase arrival-time picking performance, potentially by up to 28.8%, without extensive data curation efforts. Evaluate its impact on your specific microseismic datasets to validate its benefits.

Key insights

LaNCoR effectively mitigates label noise in seismic data by aligning feature and label representation distributions.

Principles

• Inaccurate labels significantly degrade supervised ML model performance.
• Traditional label noise reduction methods are labor-intensive and costly.

Method

The Label Noise-Contrastive Robust Learning (LaNCoR) approach aligns input waveform feature and label representation distributions in feature space to correct mislabeling during training.

In practice

• Apply LaNCoR to P-phase arrival-time picking tasks.
• Expect up to 28.8% performance improvement in seismic data processing.

Topics

• Label Noise
• Seismic Signal Processing
• P-phase Picking
• Robust Learning
• Geophysics
• Machine Learning

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video — Physics-guided deep spatiotemporal learning from video can accurately and consistently estimate coastal wave peak periods, overcoming traditional monitoring limitations.
• GNSS-FM: A Self-Supervised Foundation Model for Daily GNSS Displacement Time Series — GNSS-FM introduces self-supervised pretraining to overcome labeled data scarcity in GNSS time series analysis, outperforming supervised baselines.
• Kaggle Winners Walkthroughs: Yale/UNC-CH - Geophysical Waveform Inversion with Team F A M A S — Iterative data generation and post-optimization significantly enhance geophysical waveform inversion model performance.
• WaveDINO: Learning-Based Atmospheric Correction of Unwrapped InSAR Interferograms Validated by GNSS: Results at Laguna del Maule and Campi Flegrei Volcanoes — WaveDINO employs hybrid training and DINOv3 features for superior learning-based atmospheric correction in InSAR.
• Contrastive Learning for Seismic Horizon Tracking with Domain-Specific Priors — Fusing signal-based priors with contrastive learning enables robust seismic horizon tracking across faults.
• WaveGraphNet: Physics-Consistent Guided-Wave Damage Localization through Coupled Inverse-Forward Graph Learning — Coupling inverse graph localization with a physics-consistent forward model improves damage localization extrapolation in unseen regions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.