Auditability for Secure Machine Learning in the Cloud

Auditability for Federated Learning

MLQENABLER: Enabling Secure Machine Learning Queries over Encrypted Database in Cloud Computing

In cloud computing, the public cloud service providers (CSPs) can provide cloud storage as the primary service while providing additional machine learning (ML)-based services by using the clients'data in storage. This business model extends the border of cloud computing services and brings in new business growth possibilities. Although it is promising, the model also brings in security concerns since the public commercial cloud cannot be fully trusted. For example, the public commercial clouds may sell clients'sensitive data to the government or other companies. To address the security concerns, an immediate solution is to require clients to encrypt their datasets before outsourcing to the cloud. However, if a database is formally encrypted, then the database contains only pseudorandom numbers, making it impossible to enable ML over it. In this project, we propose MLQENABLER (ML Queries Enabler) scheme to enable secure ML queries over encrypted database in cloud storage. MLQENABLER employs an index-aid approach to achieve security and ML capability simultaneously. Our initial experiments show that MLQENABLER achieves an acceptable security level while incurring only a slight ML performance degradation.

Enhancing the Data Transmission Security in Cloud using Machine Learning

Abstract— This research outlines the cloud data security when using machine learning techniques – Random Forest, Deep Neural Networks, and Q-Learning to prevent unauthorized data transfers and leaks. The major findings point to the fact that DNN showed a higher level of prediction capabilities in comparison with Random Forest – 95% of overall accuracy as opposed to 92%. It is particularly important to consider AUC-ROC of Random Forest, which is 0.96, making it the most reliable model. However, Q-Learning appears to be less accurate with 88% yet more effective when it comes to a cumulative reward and a policy optimization – features that are vital for a changing environment of cloud servers. The findings of the research make a significant contribution to the field of cloud data security, showing the effectiveness of advanced machine learning models in terms of identifying and mitigating security breaches. This, in turn, creates the opportunity of implementation of these techniques into security frameworks for enhancing the resilience and efficiency of the latter. The recommendations for further research lie in the area of hybrid models creation, in particular, the models that would be able to utilize the positive sides of all three techniques. Moreover, the results would be more generalized with a significantly larger dataset comprising diverse cloud environments and threat situations. Finally, the investigation of the models described in a real-time  setting  and  large  cloud-based systems may be suggested for further research, given that these characteristics are essential for an effective practical deployment helping resist emerging threats. Keywords- Cloud Data Security, Machine Learning, Random Forest, Deep Neural Networks, Q-Learning

SoK: Verifiable Integrity Claims for Privacy-Preserving Federated Learning

This SoK model federated learning as an append-only transcript of submissions, admissions, aggregation, and finalization events, and formalize verifiability as a collection of integrity claims issued by clients and the aggregator, and checked by different verifier classes.

Nephology: Characterizing the Security of Base Images across Cloud Providers

AWS Security Architecture and Machine Learning for APT Detection in Cloud Environments

Cloud environments, and Amazon Web Services (AWS) in particular, host high-value data assets and mission-critical workloads that make them attractive targets for Advanced Persistent Threat (APT) actors. Because forensic investigation techniques are applied only after a breach has already been discovered, the volume and velocity of cloud-generated telemetry make proactive, automated detection capabilities essential. This paper reviews machine learning-driven anomaly detection paradigms — supervised, unsupervised, semi-supervised, and deep learning — and examines their suitability for APT detection in AWS environments. It also reviews the AWS shared-responsibility security architecture, including Identity and Access Management (IAM), encryption services, and logging and monitoring services such as AWS Cloud Trail, AWS Config, Amazon Guard Duty, Amazon Detective, and Amazon Inspector, and considers the NIST Cyber security Framework (CSF) as a governance overlay that connects these technical capabilities to organizational risk management. Drawing on this review of the peer-reviewed and primary-source literature, the paper argues that no single detection paradigm is likely sufficient on its own, and that unsupervised and semi-supervised machine learning, combined with AWS-native security services and governed by the NIST CSF, offer a more resilient conceptual basis for cloud APT defense than any single method in isolation. On this basis, the paper proposes an Integrated Cloud APT Detection and Defense Model (ICADDM) as a conceptual architecture for researchers and practitioners, maps AWS security services against the MITRE ATT&CK Cloud Matrix, and identifies the empirical validation of the model against real cloud telemetry as the principal direction for future work.