As software systems grow in complexity, source code vulnerability detection becomes crucial for software security. Existing methods, whether sequence-based or graph-based, face limitations in accurately detecting vulnerabilities. Sequence-based models often struggle with capturing code structure, while graph-based models have difficulty handling long-distance contextual relationships. To overcome these challenges, we propose a collaborative training framework that unifies a graph-based deep learning module and a semantic-rich large model module. The deep learning module, based on graph neural networks (GNNs), captures code structural information, and the large model module, leveraging pre-trained large language models (LLMs), understands code semantics. Through an iterative collaborative training mechanism, the two modules exchange information and learn from each other.Experimental results on three public datasets (Big-Vul, Reveal, and Devign) demonstrate the superiority of our approach. Compared with baseline models, our collaborative training model (CTVD) achieves significant improvements in accuracy, recall, precision, and F1-score. For example, on the Big-Vul dataset, our model’s accuracy reaches 86.5%, outperforming the deep learning module alone by 8.3% and the large model module alone by 6.4%. Compared with the latest co-training method-Vul-LMGNN, CTVD outperforms Vul-LMGNN in the DiverseVul dataset. We applied CTVD in real projects and found seven undisclosed vulnerabilities, all of which were reported and included in the CNNVD. In conclusion, our proposed collaborative training framework effectively combines the strengths of deep learning and large model modules, providing a more accurate and reliable solution for source code vulnerability detection.

Malware poses a significant threat to cyberspace security, making the classification of malware families crucial for safeguarding computer devices and information systems. Analyzing the API call behavior characteristics of malware is a key method for identifying and detecting malware. However, traditional machine learning and deep learning methods struggle to effectively capture the complex interaction patterns between API calls. In this paper, we propose a multiclass malware detection method named MaIBET, which leverages a multi-layer bidirectional Transformer encoder network to capture the intricate dependencies between API calls and identify malware families. To ensure efficient classification and acceptable accuracy, we have implemented specific improvements to the encoder network. Evaluation results show that MalBET achieves an overall performance with an average accuracy of 95.55% across four different datasets. Additionally, we conducted three sets of experiments to validate the contributions of the implemented improvements.

The use of malware for illicit cyber activities, including network attacks and information theft, poses a severe threat to cybersecurity. In comparison to traditional malware detection methods based on signature and heuristics, machine learning and deep learning-based malware detection methods demonstrate superior generalization ability. However, existing research still faces challenges such as reliance on relatively single malware features, inadequate ability to describe malware features, and overdependence on labeled data. In this paper, we propose an image-based malware classification method using self-supervised and contrastive learning, named IMCSCL. We visualize malware using opcode semantic features, and then detect malware using a contrastive learning method with improved feature encoder network. Experimental results demonstrate that IMCSCL achieves higher detection accuracy compared to supervised malware detection methods, achieving 98.85% accuracy on the Microsoft Malware Classification Challenge dataset. Fine-tuning the model using randomly selected 5% labeled samples from the training set still achieved high accuracy of 94.22%. IMCSCL exhibits superior generalization ability, faster convergence speed, and better training stability. Moreover, contrastive learning significantly reduces malware labeling costs while effectively enhancing detection performance.