We explore these existing approaches and identify common pitfalls in three main areas: (1) synthetic data generation process, (2) evaluation metrics, and (3) the final presentation of the explanation. For this purpose, we perform an empirical study to explore these pitfalls along with their ...
NF-GNN:Network Flow Graph Neural Networks for Malware Detection and Classification. (不是CFG) 本文提出CFGExpaliner,是首个针对基于GNN进行恶意软件分类(注意不是检测)的可解释性工作。CFGExplainer在CFG中提取出得到分类结果的关键子图,并对子图内节点进行重要性排序。作者通过与三个常见的GNN解释方法:GNN...
研究网络安全领域AI可解释性的工作还有Euro S&P上的《Evaluating Explanation Methods for Deep Learning in Security》,本文则主要是针对带有结构信息的图数据以及漏洞发现场景。 图神经网络在漏洞发现中的应用 源代码可以自然转化为多种有向图模型,例如: AST(抽象语法树),程序的语法结构,节点为程序语言符号,边为语法...
GInX-Eval are consistent across multipledatasets and align with human evaluation.1 I NTRODUCTIONWhile people in the f ield of explainable AI have long argued about the nature of good explanations,the community has not yet agreed on a robust collection of metrics to measure explanation correct-...
2.2Explanation via Feature Information 2.3 Multi-instance explanations through graph prototypes 3. Experiments 4. Conclusion 1.Introduction 图神经网络(Graph Neural Network), 作为深度学习领域最热门的方向之一,相关论文在各大顶会层出不穷. 但是,图神经网络的解释性问题没有得到较多的关注.图神经网络的解释性是...
GNNExplainer provides a framework to visualize what a GNN model has learnt. However, the actual explanation result may not be good enough to explain a huge graph as search space for the optimal explanation is exponentially larger than a smaller one. Instead of fitting a neural network, other ...
CGMega detects gene modules based on a model-agnostic neural network interpretation approach (Fig.1b), and these gene modules consist of two parts: i) a core subgraph consisting of the most influential pairwise relationships for the prediction of cancer gene, and ii) 15-dimensional importance ...
Graph neural networks (GNNs) have been widely used in molecular property prediction, but explaining their black-box predictions is still a challenge. Most existing explanation methods for GNNs in chemistry focus on attributing model predictions to individual nodes, edges or fragments that are not nece...
Explanation-Based Enhancement A class of approaches prompt LLMs to generate additional node explanations, descriptors or labels to enrich textual attributes. These supplement the existing text data to allow improved embedding. For example, an LLM may output research area tags for papers ...
Spatial transcriptomics (ST) data provide profiles of gene expression along with information on spatial location to be fed into a graph neural network (GNN). The choice of GNN significantly influences the quality of the generated latent embedding. For a thorough analysis of the data, principal ...