Deep learning approach to security enforcement in cloud workflow orchestration.

Area of Science:

• Computer Science
• Cybersecurity
• Cloud Computing

Background:

• Cloud workflows face significant security threats, including data breaches and insider attacks, jeopardizing sensitive data like patient health records.
• Ensuring security and data privacy in dynamic, multi-stakeholder cloud environments is a complex challenge for cloud providers.
• Existing security measures often struggle with the dynamic nature of cloud workflows and data.

Purpose of the Study:

• To propose a novel architecture and formal model for robust security enforcement in cloud workflow orchestration.
• To develop a multi-modal approach for detecting and predicting anomalies in cloud workflows.
• To implement an adaptation scheme to mitigate identified anomalies and maintain workflow performance.

Main Methods:

• A multi-modal approach combining deep learning, one-class classification, and clustering for anomaly detection and prediction.
• Unsupervised learning of static and dynamic features to reduce data dimensionality and improve task characterization.
• An adaptation scheme designed to respond to detected anomalies and minimize their impact on cloud performance.

Main Results:

• High anomaly prediction accuracy demonstrated through precision, recall, and F1 scores using a COVID-19 patient health records dataset.
• Maintained high cloud workflow execution performance after applying the adaptation strategy to detected anomalies.
• The proposed architecture effectively prevents resource wastage by accurately detecting and predicting anomalies.

Conclusions:

• The developed architecture provides a robust solution for security enforcement in cloud workflow orchestration.
• The multi-modal anomaly detection and prediction system significantly enhances the security of sensitive data in cloud environments.
• The adaptation scheme ensures resilience and optimal performance of cloud workflows even when facing security threats.