Scaling Homomorphic Applications in Deployment

TL;DR

This work tackles scalable deployment of privacy-preserving computation by integrating Fully Homomorphic Encryption ($FHE$) with a production-like movie-recommendation pipeline built on Concrete-ML. It demonstrates a complete pipeline: $FHE$-enabled inference in a containerized Flask app, orchestrated on Kubernetes, with an RL agent (PPO) learning to tune replica deployment under encryption-induced latency constraints. The study introduces an explicit MDP for deployment, robust self-healing mechanisms, and stress-testing to train a resilient controller. Key finding shows a practical replica range around $3$–$6$ balances latency and resource usage, signaling the viability of scalable, secure inference in real-world deployments.

Abstract

In this endeavor, a proof-of-concept homomorphic application is developed to determine the production readiness of encryption ecosystems. A movie recommendation app is implemented for this purpose and productionized through containerization and orchestration. By tuning deployment configurations, the computational limitations of Fully Homomorphic Encryption (FHE) are mitigated through additional infrastructure optimizations Index Terms: Reinforcement Learning, Orchestration, Homomorphic Encryption

Figures (6)

• Figure 1: General workflow of FHE inference with Concrete-ML.
• Figure 2: Learning Environment
• Figure 3: Replica Selection
• Figure 4: Reward Per Episode
• Figure 5: Latency Per Episode