Exploring Challenges in Deep Learning of Single-Station Ground Motion Records

2024-03-12

arXiv.org

AI Summary

This study identifies a “spurious correlation trap” in deep learning models for single-station ground-motion identification: models rely heavily on auxiliary cues—such as P- and S-wave arrival times—rather than learning intrinsic temporal dynamics of ground motion, severely degrading generalization (a 32% drop in event classification accuracy without auxiliary constraints). Addressing the critical gap in robust, auxiliary-independent modeling, we first systematically quantify the implicit dominance of auxiliary information. We propose a novel ablation analysis framework integrating temporal convolutional networks with attention mechanisms, complemented by Grad-CAM-based feature attribution and multivariate correlation decoupling experiments. Our results demonstrate that disentangling auxiliary dependencies significantly improves model interpretability and out-of-distribution generalization. This work advances seismological deep learning toward a genuinely time-series–centric paradigm, shifting emphasis from heuristic auxiliary features to fundamental signal dynamics.

Technology Category

• Machine Learning: Other Foundations of Machine Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Contemporary deep learning models have demonstrated promising results across various applications within seismology and earthquake engineering. These models rely primarily on utilizing ground motion records for tasks such as earthquake event classification, localization, earthquake early warning systems, and structural health monitoring. However, the extent to which these models truly extract meaningful patterns from these complex time-series signals remains underexplored. In this study, our objective is to evaluate the degree to which auxiliary information, such as seismic phase arrival times or seismic station distribution within a network, dominates the process of deep learning from ground motion records, potentially hindering its effectiveness. Our experimental results reveal a strong dependence on the highly correlated Primary (P) and Secondary (S) phase arrival times. These findings expose a critical gap in the current research landscape, highlighting the lack of robust methodologies for deep learning from single-station ground motion recordings that do not rely on auxiliary inputs.

Problem

• Assessing deep learning reliance on auxiliary seismic data
• Evaluating meaningful pattern extraction from ground motion records
• Addressing single-station learning gaps without auxiliary inputs

Innovation

• Evaluates deep learning dependence on seismic phase times
• Highlights gap in single-station motion analysis
• Proposes robust methods without auxiliary inputs