Experiences Building Enterprise-Level Privacy-Preserving Federated Learning to Power AI for Science

TL;DR

This work addresses the challenge of deploying privacy-preserving federated learning at enterprise scale for science by articulating a vision and architecture that unifies local prototyping with distributed deployment across diverse infrastructures. It proposes core capabilities—scalable local simulation, seamless sim-to-deploy transition, heterogeneous deployment support, hierarchical abstractions, and robust privacy protections—and a modular server/client architecture with an extensible, hook-based workflow. A key contribution is the architectural decoupling of FL logic from network/orchestration, plus an event-driven extensibility model and the concept of FL as a Service to simplify adoption. Together, these designs enable secure, scalable collaborative AI for data-rich scientific domains without centralized data sharing, underpinning practical impact across biomedicine, energy, climate, and astrophysics.

Abstract

Federated learning (FL) is a promising approach to enabling collaborative model training without centralized data sharing, a crucial requirement in scientific domains where data privacy, ownership, and compliance constraints are critical. However, building user-friendly enterprise-level FL frameworks that are both scalable and privacy-preserving remains challenging, especially when bridging the gap between local prototyping and distributed deployment across heterogeneous client computing infrastructures. In this paper, based on our experiences building the Advanced Privacy-Preserving Federated Learning (APPFL) framework, we present our vision for an enterprise-grade, privacy-preserving FL framework designed to scale seamlessly across computing environments. We identify several key capabilities that such a framework must provide: (1) Scalable local simulation and prototyping to accelerate experimentation and algorithm design; (2) seamless transition from simulation to deployment; (3) distributed deployment across diverse, real-world infrastructures, from personal devices to cloud clusters and HPC systems; (4) multi-level abstractions that balance ease of use and research flexibility; and (5) comprehensive privacy and security through techniques such as differential privacy, secure aggregation, robust authentication, and confidential computing. We further discuss architectural designs to realize these goals. This framework aims to bridge the gap between research prototypes and enterprise-scale deployment, enabling scalable, reliable, and privacy-preserving AI for science.

Figures (3)

• Figure 1: A typical end-to-end workflow for leveraging federated learning to train AI models. The workflow begins with the problem definition phase, where practitioners formulate the learning objective and design the model based on the available data, similar to conventional centralized training. This is followed by local simulation and algorithm development, in which practitioners verify training logic, tune hyperparameters, and prototype aggregation strategies in a simulated federated environment. Next, during the pre-deployment testing stage, the workflow is validated across distributed computing resources to ensure connectivity and configuration consistency. Finally, in the full deployment and training phase, large-scale federated training is executed across heterogeneous clients. Practitioners may iteratively revisit the simulation phase to refine configurations and improve model performance.
• Figure 2: Architectural design of the proposed FL framework. The central server coordinates multiple heterogeneous clients deployed across diverse computing environments - including HPC clusters, cloud computing (e.g, AWS, Google Cloud), and personal devices. Each client agent performs local model training with its platform-specific resource management ecosystem (e.g., Ray and Kubernetes for cloud computing, Globus Compute and Parsl for HPC or personal devices), while the server agent manages global aggregation and orchestration. Communication between clients and the server occurs over HTTP or gRPC, enabling efficient and platform-agnostic interaction.
• Figure 3: Federated Learning as a Service (FLaaS). A unified web platform enabling one-time client setup, automated experiment management, real-time monitoring, and analytics across heterogeneous clients.