跳到正文
arXiv · cs.AI · Dawei Han, Zicheng Wang, Yaoxing Ren, Yi Lv, Rongqian Wu, Fumin Wang · 4 小时前模型评分28

研究揭示磁图网络方向信息可被互易性抹除的结构边界

Reciprocity Can Erase Direction:tructural Boundaries and Controlled Utility of Magnetic Graph Networks

导读

研究发现,基于支撑集的磁图拉普拉斯在观测节点数满足 m≤k+1 时,Top-k 选择会使方向机制完全坍缩,稠密支撑还会给互易弧比例设下下界。在 72 次固定神经网络拟合中,支撑相位在单向环上测试 AUROC 达 1.0000,但在互易环和完全支撑上仅 0.5000,而连续分数相位与实数有向对照模型全程保持 1.0000。

分类置信度

中 · 待人工复核

  • 只有单一媒体来源,需要人工复核。

来源级别:媒体与个人

事件类型:研究发布

证据摘录

  • “Magnetic graph networks encode direction through phase, but a learned graph can erase the intended directional mechanism before prediction.”
  • “These are oracle-direction mechanism tests, not evidence of learned clinical direction.”
  • “It provides a precise circumstance in which magnetic phase preserves useful information, without asserting universal or clinical superiority.”

来源:arXiv · cs.AI · arxiv.org