研究人员开发了ZAYAN，一个新颖的自监督框架，旨在改进表格遥感数据的表示学习。这种以特征为中心的对比方法在特征级别上运行，无需显式的锚点或类别标签。该框架包括用于预训练特征嵌入的ZAYAN-CL和一个利用这些嵌入进行下游分类任务的Transformer ZAYAN-T。ZAYAN在各种数据集上展示了卓越的准确性和鲁棒性，尤其是在标签稀缺和分布偏移的条件下。 AI
影响 引入了一种从表格遥感数据学习的新方法，有望提高环境科学应用中的准确性和鲁棒性。
排序理由 这是一篇描述表格数据新框架的研究论文。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2604.27606v1 Announce Type: cross Abstract: Learning informative representations from tabular data in remote sensing and environmental science is challenging due to heterogeneity, scarce labels, and redundancy among features. We present ZAYAN (Zero-Anchor dYnamic feAture eN…
Learning informative representations from tabular data in remote sensing and environmental science is challenging due to heterogeneity, scarce labels, and redundancy among features. We present ZAYAN (Zero-Anchor dYnamic feAture eNcoding), a self-supervised, feature-centric contra…
Learning informative representations from tabular data in remote sensing and environmental science is challenging due to heterogeneity, scarce labels, and redundancy among features. We present ZAYAN (Zero-Anchor dYnamic feAture eNcoding), a self-supervised, feature-centric contra…