GIAT: A Geologically-Informed Attention Transformer for Lithology Identification
This addresses the need for more accurate and interpretable deep learning models in geoscience applications, particularly for subsurface resource evaluation.
The paper tackled the problem of lithology identification from well logs by proposing the Geologically-Informed Attention Transformer (GIAT), which integrates geological priors into the Transformer's attention mechanism, achieving state-of-the-art accuracy up to 95.4% on two datasets.
Accurate lithology identification from well logs is crucial for subsurface resource evaluation. Although Transformer-based models excel at sequence modeling, their "black-box" nature and lack of geological guidance limit their performance and trustworthiness. To overcome these limitations, this letter proposes the Geologically-Informed Attention Transformer (GIAT), a novel framework that deeply fuses data-driven geological priors with the Transformer's attention mechanism. The core of GIAT is a new attention-biasing mechanism. We repurpose Category-Wise Sequence Correlation (CSC) filters to generate a geologically-informed relational matrix, which is injected into the self-attention calculation to explicitly guide the model toward geologically coherent patterns. On two challenging datasets, GIAT achieves state-of-the-art performance with an accuracy of up to 95.4%, significantly outperforming existing models. More importantly, GIAT demonstrates exceptional interpretation faithfulness under input perturbations and generates geologically coherent predictions. Our work presents a new paradigm for building more accurate, reliable, and interpretable deep learning models for geoscience applications.