从地球气候到电网,从社会群体到人类大脑,复杂的系统都可能发生突变,例如全球迅速变暖、停电或疾病。不仅要能预测这些事件的来临,还要能预测它们的位置,这对减轻持久后果至关重要。现在,刘子嘉及其同事提出了一种机器学习框架,可以预测大规模网络中临界转换的确切位置(Z. Liu et al. Phys. Rev. X 14, 031009; 2024)。图同构网络的各层处理每个节点的时间序列数据,提取它们各自的特征。然后将这些信息整合到整个图中,并由门控递归单元层读取,门控递归单元层负责识别重复出现的节点特征。
期刊介绍:
Nature Physics is dedicated to publishing top-tier original research in physics with a fair and rigorous review process. It provides high visibility and access to a broad readership, maintaining high standards in copy editing and production, ensuring rapid publication, and maintaining independence from academic societies and other vested interests.
The journal presents two main research paper formats: Letters and Articles. Alongside primary research, Nature Physics serves as a central source for valuable information within the physics community through Review Articles, News & Views, Research Highlights covering crucial developments across the physics literature, Commentaries, Book Reviews, and Correspondence.