Smart Contract Vulnerability Detection Model Based on Multi-Task Learning

Area of Science:

• Computer Science
• Software Engineering
• Artificial Intelligence

Background:

• Smart contract security is crucial, yet current vulnerability detection methods lack efficiency and type identification capabilities.
• Existing approaches often struggle with scalability and precise vulnerability classification.

Purpose of the Study:

• To develop an efficient and scalable smart contract vulnerability detection model.
• To enhance the model's ability to not only detect but also recognize specific types of vulnerabilities.

Main Methods:

• A multi-task learning model with a hard-sharing design was developed.
• A bottom sharing layer learns contract semantics using word/positional embeddings and an attention-based neural network.
• Task-specific layers, utilizing convolutional neural networks, perform classification for each task.

Main Results:

• The multi-task model demonstrated improved capability in identifying and recognizing specific vulnerability types.
• Experimental results confirmed the model's effectiveness in detecting and classifying three types of vulnerabilities.
• The multi-task approach proved more efficient in terms of time, computation, and storage compared to single-task models.

Conclusions:

• The proposed multi-task learning model offers a superior solution for smart contract vulnerability detection and classification.
• This approach enhances detection accuracy and provides type recognition, addressing limitations of existing methods.
• The model presents a more cost-effective and efficient solution for smart contract security analysis.