SPHINX：基于超图推理网络的结构预测
材料科学
2025-10-28 v2
摘要
在众多真实系统中，人们广泛认识到更高阶关系的重要性。然而，对它们进行标注往往是繁琐且有时不可能的。因此，当前的数据建模方法要么完全忽略更高阶相互作用，要么将其简化为二元连接。为了促进更高阶处理，即使超图结构不可用，我们引入了结构预测使用超图推理网络（SPHINX），这是一种模型，能够从最终节点级信号中以无监督方式推断潜在超图结构。该模型由一个软的、可微的聚类方法用于顺序预测每个超边上节点概率分布，并由一个采样算法将其转换为显式的超图结构。我们表明，近期进展的 -子集采样代表了一种合适的工具，用于产生离散的超图结构，解决了先前作品所表现的训练不稳定性问题。 resulting model can generate the higher-order structure necessary for any modern hypergraph neural network, facilitating the capture of higher-order interaction in domains where annotating them is difficult. Through extensive ablation studies and experiments conducted on two challenging datasets for trajectory prediction, we demonstrate that our model is capable of inferring suitable latent hypergraphs, that are interpretable and enhance the final performance.
引用
@article{arxiv.2410.03206,
title = {A simple method to find temporal overlap between THz and X-ray pulses using X-ray-induced carrier dynamics in semiconductors},
author = {Yuya Kubota and Takeshi Suzuki and Shigeki Owada and Kenji Tamasaku and Hitoshi Osawa and Tadashi Togashi and Kozo Okazaki and Makina Yabashi},
journal= {arXiv preprint arXiv:2410.03206},
year = {2025}
}