An Enhanced Method for Virtual Machine Anomaly Detectors Based on Hybrid Fine-Tuning Strategy
Virtual machine anomaly detectors are faced with the problem of sparse training data samples in realistic scenarios.To address the issue of accurate detection of virtual machine anomalies under such sparse sample condi-tions,a virtual machine detection enhancement method based on a hybrid fine-tuning strategy is proposed in this article.Firstly,by using a large language model and based on prompt engineering,the collected virtual machine(VM)escape samples are upgraded,and the enhanced samples are obtained..Secondly,a hybrid fine-tuning strategy combining LoRA and PreFix is are used to enhance the collected VM escape samples and obtain enhanced samples.Subsequently,the enhanced samples are utilized to fine-tune the pre-trained large model,ChatGLM,thereby cre-ating a dedicated model for generating VM escape data.Finally,the generated samples from the dedicated model are leveraged to augment the dataset and improve the performance of the learning-based VM anomaly detector.Ex-perimental validation demonstrates that this method achieves significant improvements in VM escape detection ac-curacy while effectively reducing the false alarm rate.
万方数据
维普
