Probing Graph Neural Network Activation Patterns Through Graph Topology
Department of Computer Science, University of Geneva, Geneva, Switzerland(日内瓦大学计算机科学系);Research Institute for Statistics and Information Science, Geneva, Switzerland(日内瓦统计与信息科学研究所);School of Engineering and Applied Sciences, Harvard University(哈佛大学工程与应用科学学院)
Floriano Tori, Lorenzo Bini, Marco Sorbi, Stéphane Marchand-Maillet, Vincent Ginis
图上的曲率概念为图拓扑提供了理论描述,突显了瓶颈和更密集连接的区域。图神经网络中信息传递范式的副产品,如过平滑和过挤压,已被归因于这些区域。然而,图拓扑与GNNs学习偏好之间的相互作用仍然不明确。通过Massive Activations,即图变换器中的极端边激活值,我们探测这种对应关系。在合成图和分子基准上,我们的发现表明,MAs并不优先集中在曲率极值处,尽管它们理论上与信息流相关。在Long Range Graph Benchmark上,我们识别出一种系统性的曲率转移:全局注意力机制加剧了拓扑瓶颈,大幅增加了负曲率的普遍性。我们的工作将曲率重新界定为一种诊断探针,用于理解何时以及为何图学习失败。
英文摘要:
Curvature notions on graphs provide a theoretical description of graph topology, highlighting bottlenecks and denser connected regions. Artifacts of the message passing paradigm in Graph Neural Networks, such as oversmoothing and oversquashing, have been attributed to these regions. However, it remains unclear how the topology of a graph interacts with the learned preferences of GNNs. Through Massive Activations, which correspond to extreme edge activation values in Graph Transformers, we probe this correspondence. Our findings on synthetic graphs and molecular benchmarks reveal that MAs do not preferentially concentrate on curvature extremes, despite their theoretical link to information flow. On the Long Range Graph Benchmark, we identify a systemic \textit{curvature shift}: global attention mechanisms exacerbate topological bottlenecks, drastically increasing the prevalence of negative curvature. Our work reframes curvature as a diagnostic probe for understanding when and why graph learning fails.
Detection Without Correction: A Robust Asymmetry in Activation-Based Hallucination Probing
Department of Computer Science and Engineering, Indian Institute of Technology Patna, India(印度理工学院帕纳分校计算机科学与工程系);Department of Computer Science, Rajarshi School of Management and Technology, Varanasi, India(印度瓦拉纳西 Rajarshi 管理与技术学校计算机科学系);Department of Electronics and Communication Engineering, Jaypee Institute of Information Technology, Noida, India(印度诺伊达 Jaypee 信息科技学院电子与通信工程系)
Experimental Observations of the Topology of Convolutional Neural Network Activations
Pacific Northwest National Laboratory(太平洋西北国家实验室);Scientific Computing and Imaging (SCI) Institute and School of Computing, University of Utah(犹他大学科学计算与成像研究所和计算学院)
拓扑数据分析(Topological Data Analysis,TDA)是计算数学的一个分支,是代数拓扑学和数据科学的桥梁,它提供了复杂结构的紧凑,抗噪的表示。深度神经网络(DNN)学习与模型架构定义的一系列转换相关的数百万个参数,导致输入数据的高维、难以解释的内部表示。随着DNN在我们社会的多个领域变得越来越普遍,人们越来越认识到需要数学方法来帮助分析师,研究人员和从业者理解和解释这些模型的内部表示与最终分类的关系。在本文中,我们采用尖端技术从TDA的目标是深入了解用于图像分类的卷积神经网络的可解释性。我们使用两种常见的TDA方法来探索几种方法来模拟隐藏层激活的高维点云,并提供实验证据,这些点云捕捉有价值的结构信息模型的过程。首先,我们证明了基于持久同源性的距离度量可以用来量化层之间有意义的差异,我们讨论了这些距离在更广泛的背景下,现有的代表性相似性度量的神经网络可解释性。其次,我们表明,映射图可以提供语义洞察这些模型如何组织层次类知识在每一层。这些观察结果表明,TDA是一个有用的工具,可以帮助深度学习从业者解锁其模型的隐藏结构。
Socioeconomic correlations of urban patterns inferred from aerial images: interpreting activation maps of Convolutional Neural Networks
Univ Lyon(里昂大学);Inria(法国国家信息与自动化研究所);CNRS(法国国家科学研究中心);ENS de Lyon(里昂高等师范学院);Université Claude Bernard Lyon 1(里昂第一大学克洛德·贝尔纳大学);Central European University(中欧大学)