通过子图顺序嵌入空间,学习模型级别的图形神经网络的解释
Li Liu1, Pengyu Wan2, Feiyan Zhang2
1Chongqing Key Laboratory of Computational Intelligence, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China; School of Computer Science and Technology, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China; Key Laboratory of Cyberspace Big Data Intelligent Security, Ministry of Education, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China.
本研究介绍了MOSE,这是一种新的图形神经网络 (GNN) 解释器,可以生成可靠的图形模式,以更好地理解模型. MOSE确保了解释的典型性和效率,提高了GNN的解释性.
科学领域:
- 人工智能的人工智能
- 机器学习 机器学习
- 图形神经网络的神经网络
背景情况:
- 模型级图形神经网络 (GNN) 解释器识别了用于类预测的关键图形模式.
- 现有的方法往往缺乏约束,产生不可靠和非典型的模式,阻碍解释质量.
研究的目的:
- 提出MOSE (通过子图顺序嵌入空间进行MOdel级解释),这是一个用于生成可靠和典型图形模式的新解释器.
- 通过将MOSE扩展到节点分类任务来增强GNN可解释性的概括性.
主要方法:
- MOSE使用图形编码器创建一个嵌入空间,保留子图关系.
- 使用贪抽样的得分函数产生受约束的,可靠的图形模式候选者.
- 解释候选人是根据GNN预测的概率来选择的.
主要成果:
- MOSE在多个指标上表现出有效性,包括预测准确性,模型实用性和效率.
- 在合成和现实数据集上的实验验验证了MOSE的性能.
- 将MOSE扩展到节点分类显示了增强的泛化.
结论:
- MOSE提供了一种简单而有效的方法,用于生成可靠和典型的模型级GNN解释.
- 该方法通过结合子图顺序嵌入来解决当前解释器的局限性.
- MOSE为改善GNN的解释性和适用性提供了一个有希望的方向.
相关概念视频
Vector Algebra: Graphical Method
We use the laws of geometry to construct resultant vectors, followed by trigonometry to find vector magnitudes and directions. For a geometric construction of the sum of two vectors in a plane, we follow the parallelogram rule. Suppose two vectors are at arbitrary positions. Translate either one of...
Sequence Networks of Rotating Machines
Zero-sequence current induces a voltage drop across the generator's neutral impedance and other...
Neural Circuits
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
Circuit Terminology
A circuit, on the other hand, is also an interconnected system of electrical elements but must contain one or more closed paths.
Block Diagram Reduction
The first step in this process is the identification and relocation of a branch point. A branch point, where a...
Woodward–Hoffmann Selection Rules and Microscopic Reversibility


