To address the challenges of diverse landslide morphologies and strong background interference in post-earthquake mountainous regions, where a single model suffers from limited adaptability, this paper proposes a dual-model ensemble method for remote sensing landslide identification based on Swin Transformer. The method employs Swin Transformer as a shared backbone network to reduce computational redundancy, integrates the core enhancement modules from SCPD-Deeplabv3+ and LSMFormer to simultaneously capture the global structure of complex landslides and fine-scale boundary details of small landslides, and reuses the MSAD decoder for deep feature fusion to achieve pixel-level segmentation. Experimental results show that the ensemble model achieves a mean intersection over union (mIoU) of 91.88%, with precision, recall, and F1-score of 94.37%, 96.11%, and 94.78%, respectively. The proposed method outperforms each individual model, effectively reducing false positives and missed detections, while balancing the contour accuracy of large-scale landslides and the detail precision of small-scale landslides.
Xiaoyu Fan, Xiaobin Li, T. Ma et al.· International Conference on...· 0 citations
Foundation Models for Electronic Health Records (FEMRs) are pretrained on large-scale structured patient data, enabling them to convert longitudinal patient trajectories into generalizable representations for diverse clinical prediction tasks. Despite their effectiveness, FEMRs remain black-box models, raising concerns about bias, interpretability, and clinical trust. To address this, we propose the first token-level explainability approach for FEMRs. We train a Transformer-based surrogate model on input-output pairs from the FEMR across two prediction tasks, approximating its behavior while preserving temporal dynamics. We identify the most influential tokens, providing insights into how FEMRs leverage different aspects of patient history for predictions. To evaluate clinical relevance, we introduce a novel clinical alignment metric that quantifies the correspondence between the surrogate model's key tokens and clinically validated features. Our results demonstrate that the surrogate closely approximates FEMR predictions and that token-level explanations align well with clinical knowledge, offering a practical framework for interpretable and trustworthy clinical AI.
Jie Huang, Peng Yin, Zihan Xu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
