Malicious JavaScript Detection via AST and Graph Convolutional Neural Networks

The patent replaces traditional supervised machine learning mechanisms with deep machine learning models (specifically graph convolutional neural networks) that automatically learn representations from AST structures. This substitution eliminates the need for manual feature engineering and traditional ML pipelines, resulting in more accurate detection with reduced computational overhead through optimized deep learning inference.

The patent changes the fundamental parameters of the detection system by transitioning from traditional ML algorithms to deep learning models that operate on transformed AST representations. This parameter change enables the system to handle sparse and obfuscated scripts more effectively while reducing the resource intensity associated with traditional supervised learning approaches.

The deep learning model performs self-service by automatically learning optimal feature representations from the AST structures during training. This eliminates the need for external feature engineering efforts, as the model autonomously identifies and extracts relevant patterns from the input data, significantly reducing development time and complexity.

The system performs preliminary action by pre-processing scripts into standardized AST representations before feeding them to the deep learning model. This preliminary transformation simplifies the subsequent classification task and eliminates the need for complex feature engineering, as the AST structure provides a normalized representation that the model can process directly.