Publication Date
Spring 2026
Degree Type
Master's Project
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
First Advisor
Teng Moh
Second Advisor
Melody Moh
Third Advisor
Amith Kamath-Belman
Keywords
predictive autoscaling, differential privacy, federated learning, homomorphic encryption, membership inference, tail latency
Abstract
We report the research findings of studies done on privacy-preserving predictive autoscaling for cloud environments. Essentially, we ask how can we use a predictive autoscaling service that protects client data while staying within the real-time latency constraints? The first study talks about how we can use differential privacy during model training. We use stochastic gradient descent combined with federated learning on the client side, then evaluate membership inference attacks, heterogenous clients in federated learning, and privacy guarantees. The second contribution uses anomaly detection, role based access control, and a homomorphic encryption boundary during model runtime. Finally, we tune the final layer of the prediction head under practical tail latency metrics. These three studies together help us achieve a meaningful result in implementing a predictive autoscaler that preserves data privacy.
Recommended Citation
Chuang, Alan, "A Study of Privacy-Preserving Predictive Autoscaling in Cloud Environments" (2026). Master's Projects. 1830.
DOI: https://doi.org/10.31979/etd.dsyn-ngr8
https://scholarworks.sjsu.edu/etd_projects/1830
