王正文 王俊峰



摘 要 ???:未知惡意流量是網絡安全的重大安全挑戰,對未知惡意流量的分類能夠增強網絡威脅識別能力,指導網絡防御策略.未知惡意流量由于缺乏樣本,無法滿足現有的深度學習方法對大量數據的需要.本文提出了一種基于生成式零樣本學習的未知惡意流量分類方法.從原始的網絡流量中提取出關鍵的惡意流量信息并轉化為二維圖像,提出將惡意流量的屬性信息作為輔助語義信息,利用條件生成對抗網絡生成類別樣本.同時,本文還添加了類級別的對比學習網絡,使得生成的類別樣本質量更高并且更具有類間區分度.實驗結果表明,該方法在未知惡意流量分類問題上平均準確率能夠達到90%以上,具有較高的應用價值.
關鍵詞 :?零樣本學習; 未知惡意流量; 生成對抗網絡; 對比學習
中圖分類號 :TP39308 文獻標識碼 :A DOI : ?10.19907/j.0490-6756.2023.042003
Unknown malicious traffic classification method ?based on generative Zero-shot learning
WANG Zheng-Wen, WANG Jun-Feng
(College of Computer Science, Sichuan University, Chengdu 610065, China)
Unknown malicious traffic is a major security challenge for network security, and the classification of unknown malicious traffic can enhance network threat identification and guide network defense strategies. the lack of unknown malicious samples cannot meet the need of existing deep learning methods for large amount of data. To address this problem, we propose a generative zero-shot learning based method for classifying unknown malicious traffic. The key malicious traffic information is extracted from the original network traffic and transformed into two-dimensional images, and the attribute information of malicious traffic is proposed as auxiliary semantic information to generate class samples using conditional adversarial networks. In this paper, we also add a class-level comparative learning network to generate class samples with higher quality and more differentiation between classes. The experimental results show that the average accuracy of this method can reach more than 90% in the classification problem of unknown malicious traffic. It has high application value.
Zero-shot learning; Unknown malicious traffic; Generative adversarial network; Contrastive ?learning
1 引 言
近年來,計算機網絡技術給人們的生活帶來了巨大便利,同時,也產生了一系列的安全威脅.現有的入侵檢測方法大多依賴于歷史的流量數據集,通過提取歷史流量的特征作為惡意流量的檢測指導.在實際檢測中,如果惡意流量的種類曾經出現在歷史流量數據集中,那么通過先前所獲得的流量特征,對后續的流量可以獲得很好的檢測結果.例如Gu等人 ?[1]通過統計包的大小、協議處理時間和到達時間三種數據,來對數據包進行快速分類.Alshammari等人 ?[2]從加密流量中提取了20種流統計特征,然后利用機……