Component CFG based Android Vetting System by Applying CapsGNN

Android is an event-based/component-based system. By triggering different events, the control flow can be various, plus building component control flow graph could cuasue computational challenge. The graph we are using is based on highly precise inter-procedural control flow graph (API-call nodes only) of one app's components.

The development of this vetting system's capsule graph neural netowrk is based on:

Capsule Graph Neural Network. Zhang Xinyi, Lihui Chen. ICLR, 2019. [CapsGNN]

Package Version

networkx    2.4
numpy       1.16.2
scipy       1.4.1
tensorflow  1.12.2

Note:
It can run on tensorflow 1.15 and the latest version of the other packages when this thesis was submitted.
Beaware though, of CuDNN errors that might surface, fix would be point cuDNN installation to correct cuDNN version directory.

Usage

Graph embedding

1. Sample graph data is in graph_dataset/APKICFG. Each graph is a combination one apk's all component control flow graph. To geneate more graphs, check icfg_generator.

2. To processing graph embedding:

  $ python dataset_utils/preprocessing.py

Training model

All the hyperparameters can be set in config.py. Training a model by using default configs:

  $ python main.py

Training a model with specific epochs:

  $ python main.py --epochs 10

(Would require amandroid based jar file to extract API sequence graphs.)