arXiv cs.AI· Laura Laurenti, David Liu, Brian Kulis, Men-Andrin Meier, Christopher Johnson·· 3 小时前AI 评分37
地震基础模型不只是“高级特征提取器”吗？
Are Seismic Foundation Models More Than Glorified Feature Extractors?
AI 导读
一项 arXiv 研究训练了 9 个共享 Mamba 编码器-解码器、基于 3M 窗口语料的地震波形模型，用不同自监督目标对比评估。结果显示改变学习目标只会重新分配下游能力而非提升：加掩码压力后少样本 P 波拾取精度提高 6.9 点，但少样本分类准确率下降 9.6 点。研究认为按语言与视觉领域的标准，这些模型仍是特征提取器而非基础模型，关键缺失是跨事件尺度的学习目标。
正文 · 原文
Abstract:Earthquake sequences are stochastic on two separated timescales: the seismic waves of any individual earthquake propagate over seconds to minutes, while the earthquakes themselves arrive as a sparse, self-exciting occurrence process, with inter-event intervals from seconds to months. Respecting both scales in one training objective is the open challenge. Foundation models (FMs) for seismic waveforms instead import their learning targets from audio and vision, where every target is computed from a single fixed-length window. Such a window holds one realization of the wave propagation (an earthquake trace), never the occurrence process that decides when earthquakes happen. We develop nine model designs sharing one Mamba encoder-decoder, trained on a fixed corpus of 3M windows, varying the self-supervised objective, quantization bottleneck and mixer; the two key comparisons are architecture-matched, varying the objective alone. Each checkpoint is evaluated frozen on three downstream tasks. Varying the target redistributes downstream skill rather than raising it, and no checkpoint wins everywhere. Adding masking pressure to our strongest classifier re-specializes it rather than making it a generalist: it gains 6.9 points of few-shot P-wave picking precision and loses 9.6 points of few-shot classification accuracy. By the standard imported from language and vision, these models are feature extractors rather than FMs. Controls on capacity, architecture, and domain proximity do not explain why skill moves rather than rises. Against task-specific baselines the advantage is regime-dependent: our checkpoints beat them few-shot on classification and ground-motion prediction, and trail them on full data. We argue the binding constraint is the target, and that settling the question requires a learning target defined across events, the timescale that makes seismicity stochastic in the first place.
| Comments: | NEURIPS STODY WORKSHOP 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.38003 [cs.LG] |
| (or arXiv:2610.38003v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.38003 arXiv-issued DOI via DataCite |
Submission history
From: Laura Laurenti [view email]
[v1]
Tue, 27 Oct 2026 21:54:33 UTC (423 KB)
来源：arXiv cs.AI · arxiv.org