Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Learning in Practice: Reflections and Projections

Katharine Daly, Hubert Eichner, Peter Kairouz, H. Brendan McMahan, Daniel Ramage, Zheng Xu

TL;DR

Federated Learning in Practice surveys the evolution, practice, and challenges of privacy-preserving collaborative learning, arguing for a privacy-centered redefinition of FL and the use of confidential cloud computation with TEEs to enable externally verifiable guarantees. It reviews industry deployments, secure aggregation, and differential privacy techniques, highlighting both practical successes and remaining gaps in verifiability and scalability. The authors outline a future architecture of confidential federated computations to support large foundation models while maintaining data minimization and user control, acknowledging risks such as TEE side channels and policy complexity. Overall, the work provides a roadmap for advancing FL from production systems toward privacy-verified, scalable, and open Ecosystem-enabled deployments.

Abstract

Federated Learning (FL) is a machine learning technique that enables multiple entities to collaboratively learn a shared model without exchanging their local data. Over the past decade, FL systems have achieved substantial progress, scaling to millions of devices across various learning domains while offering meaningful differential privacy (DP) guarantees. Production systems from organizations like Google, Apple, and Meta demonstrate the real-world applicability of FL. However, key challenges remain, including verifying server-side DP guarantees and coordinating training across heterogeneous devices, limiting broader adoption. Additionally, emerging trends such as large (multi-modal) models and blurred lines between training, inference, and personalization challenge traditional FL frameworks. In response, we propose a redefined FL framework that prioritizes privacy principles rather than rigid definitions. We also chart a path forward by leveraging trusted execution environments and open-source ecosystems to address these challenges and facilitate future advancements in FL.

Federated Learning in Practice: Reflections and Projections

TL;DR

Federated Learning in Practice surveys the evolution, practice, and challenges of privacy-preserving collaborative learning, arguing for a privacy-centered redefinition of FL and the use of confidential cloud computation with TEEs to enable externally verifiable guarantees. It reviews industry deployments, secure aggregation, and differential privacy techniques, highlighting both practical successes and remaining gaps in verifiability and scalability. The authors outline a future architecture of confidential federated computations to support large foundation models while maintaining data minimization and user control, acknowledging risks such as TEE side channels and policy complexity. Overall, the work provides a roadmap for advancing FL from production systems toward privacy-verified, scalable, and open Ecosystem-enabled deployments.

Abstract

Federated Learning (FL) is a machine learning technique that enables multiple entities to collaboratively learn a shared model without exchanging their local data. Over the past decade, FL systems have achieved substantial progress, scaling to millions of devices across various learning domains while offering meaningful differential privacy (DP) guarantees. Production systems from organizations like Google, Apple, and Meta demonstrate the real-world applicability of FL. However, key challenges remain, including verifying server-side DP guarantees and coordinating training across heterogeneous devices, limiting broader adoption. Additionally, emerging trends such as large (multi-modal) models and blurred lines between training, inference, and personalization challenge traditional FL frameworks. In response, we propose a redefined FL framework that prioritizes privacy principles rather than rigid definitions. We also chart a path forward by leveraging trusted execution environments and open-source ecosystems to address these challenges and facilitate future advancements in FL.

Paper Structure

This paper contains 12 sections, 1 figure, 1 table.

Table of Contents

  1. Evolution of Federated Learning
  2. Advances in Federated Learning in Practice
  3. Applications
  4. Systems
  5. Privacy
  6. Algorithms
  7. Challenges and Opportunities
  8. Scaling to Large Foundation Models
  9. Verifying Server-side Privacy Guarantees
  10. Addressing System Challenges
  11. A Path to the Future
  12. Conclusion

Figures (1)

  • Figure 1: A prototype architecture for using confidential federated computations to train large models. In contrast to traditional cross-device federated learning, devices upload (re-processed and encrypted data, and an iterative training process is performed on the server. An orchestrator is responsible for passing encrypted blobs amongst storage locations and components running in TEEs. The ledger enforces that workload specific transformations adhere to the access policy associated with data uploads.