Multi-modal Learning for Real-time Bot Detection in Cloud-based Applications
Milavkumar Shah,Swapnil Patil,Advait Patel · 2021
In the rapidly evolving landscape of cloud-based applications, the prevalence and sophistication of bot attacks pose significant threats to security and operational integrity. Traditional bot detection methods often fall short when confronted with adversarial tactics designed to deceive machine learning models. This study introduces an adversarially robust multi-modal learning framework tailored for real-time bot detection in cloud environments. Leveraging publicly available datasets such as the Bot-IoT and UNSW-NB15, our approach integrates diverse data sources, including network traffic patterns and user behavior metrics, to enhance detection accuracy and resilience. We implement advanced adversarial defense mechanisms, including adversarial training and defensive distillation, to safeguard the model against various attack vectors. Comprehensive experiments demonstrate that our multi-modal system not only outperforms conventional single-modal approaches but also maintains high performance levels under adversarial conditions. These findings underscore the potential of robust multi-modal learning in fortifying cloud security infrastructures against increasingly evasive bot threats. This work paves the way for more secure and reliable cloud-based applications by addressing the critical challenge of adversarial robustness in real-time bot detection.