Detect Advanced Persistent Threat In Graph-Level Using Competitive AutoEncoder
Open Access
- 16 June 2023
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 28-34
Abstract
APT(Advanced Persistent Threat) are known for their stealth, persistence, and sophistication, often carried out by skilled and well-funded attackers. To detect and investigate APT attacks at the host-level, scholars have extensively studied the provenance graph, which preserves the complete attack context. However, existing methods have impractical requirements for data collection and threshold setting, severely impacting their real-world performance. In this paper, we propose GCA (Graph Competitive Autoencoder), a GNN-based graph-level anomaly detection system. We first extract entities and their interactions from host logs to construct a provenance graph. Subsequently, we utilize GNN (Graph Neural Network) as a graph encoder to obtain graph-level embeddings. In the anomaly detection part, we use the competitive autoencoder and mutual information maximization to identify the attack graph. With this model, we can conduct graph-level anomaly detection even when the training data is partially contaminated, without manually setting the threshold. We deploy and evaluate our model on the public dataset StreamSpot. Our results show that GCA outperforms existing methods in terms of accuracy and practicality.Keywords
This publication has 15 references indexed in Scilit:
- Tactical Provenance Analysis for Endpoint Detection and Response Systems2020
- Modeling and detection of the multi-stages of Advanced Persistent Threats attacks based on semi-supervised learning and complex networks characteristicsFuture Generation Computer Systems, 2020
- You Are What You Do: Hunting Stealthy Malware via Data Provenance Analysis2020
- POIROT2019
- HOLMES: Real-Time APT Detection through Correlation of Suspicious Information Flows2019
- A Survey on Advanced Persistent Threats: Techniques, Solutions, Challenges, and Research OpportunitiesIEEE Communications Surveys & Tutorials, 2019
- Fast Memory-efficient Anomaly Detection in Streaming Heterogeneous Graphs2016
- node2vec2016
- DeepWalk2014
- Relational learning via collective matrix factorization2008