DID-eFed: Facilitating Federated Learning as a Service with Decentralized Identities

Jiahui Geng, Neel Kanwal, Martin Gilje Jaatun, Chunming Rong

Abstract

We have entered the era of big data, and it is considered to be the "fuel" for the flourishing of artificial intelligence applications. The enactment of the EU General Data Protection Regulation (GDPR) raises concerns about individuals' privacy in big data. Federated learning (FL) emerges as a functional solution that can help build high-performance models shared among multiple parties while still complying with user privacy and data confidentiality requirements. Although FL has been intensively studied and used in real applications, there is still limited research related to its prospects and applications as a FLaaS (Federated Learning as a Service) to interested 3rd parties. In this paper, we present a FLaaS system: DID-eFed, where FL is facilitated by decentralized identities (DID) and a smart contract. DID enables a more flexible and credible decentralized access management in our system, while the smart contract offers a frictionless and less error-prone process. We describe particularly the scenario where our DID-eFed enables the FLaaS among hospitals and research institutions.

Introduction

In the modern era of artificial intelligence, data is the essential fuel for innovation. However, the dual challenges of strict privacy regulations (such as GDPR and HIPAA) and the existence of "data silos"—where sensitive information is trapped within individual organizations—hinder the development of robust AI models. While Federated Learning (FL) offers a path forward by allowing models to be trained locally without sharing raw data, a standardized infrastructure for providing this as a service (FLaaS) to third parties has been missing.

Traditional Machine Learning as a Service (MLaaS) models typically require data to be uploaded to a central cloud, resulting in a loss of data sovereignty. DID-eFed addresses this gap by creating a secure, automated framework that allows external researchers to access the utility of private data silos without the data ever leaving its original owner.

System Architecture and Methodology

The DID-eFed framework integrates three core technologies to facilitate secure cross-organizational collaboration:

  1. Federated Learning (FL): Enables collaborative model training without direct data exchange.
  2. Decentralized Identity (DID): A blockchain-based, user-centric identity system. Instead of relying on a central authority, users present "verifiable claims" (e.g., professional credentials) to gain access, which mitigates Sybil attacks and enhances user control.
  3. Smart Contracts: These self-executing scripts on the blockchain act as automated gatekeepers, verifying DID claims against data owner policies to grant or deny access to the FL environment.

To interact with the system, users utilize two distinct API layers: a Data API for viewing metadata and customizing filters, and a Model API for defining neural network structures and training parameters.

Identity Verification Workflows

A critical component of the DID-eFed architecture is the method of identity verification. The authors analyze two primary workflows for looking up decentralized identifiers.

Figure 1: The proposed 8-step workflow where the smart contract initiates the DID lookup.

In the preferred workflow (Figure 1), the smart contract initiates the identity lookup. This method is argued to be more secure as it prevents users from potentially tampering with their own identity documents before verification.

Figure 2: An alternative workflow where the user initiates the DID lookup.

Conversely, Figure 2 illustrates an alternative where the user initiates the lookup. The paper provides a qualitative comparison, suggesting that the smart-contract-initiated approach offers better protection against identity tampering and Denial of Service (DoS) attacks.

Key Insights

  • Claim-Based Identity: Moving away from traditional Role-Based Access Control (RBAC), DID-eFed utilizes claim-based identity. This is more flexible for cross-organizational scenarios, such as a researcher from one hospital needing to be recognized by another institution.
  • Frictionless Authorization: By automating the verification process through smart contracts, the system eliminates the manual review phases and paperwork typically associated with medical data research, reducing human error.
  • High Availability: The design includes "standby node" configurations for both clients and servers, ensuring that the federated training process remains resilient even if individual institutional servers go offline.

Conclusion and Future Work

DID-eFed presents a conceptual framework for a portable, containerized FLaaS system using Docker. While it provides a robust blueprint for secure data utility, several challenges remain for future exploration:

  • Resource Scheduling: Managing simultaneous training tasks across multiple institutions to avoid CPU/GPU bottlenecks.
  • Privacy-Accuracy Balance: Optimizing the trade-off between advanced encryption (like Homomorphic Encryption) and model performance.
  • Incentive Mechanisms: Developing blockchain-based token systems to reward data owners for providing high-quality datasets.

Summary updated

Abstract

We have entered the era of big data, and it is considered to be the "fuel" for the flourishing of artificial intelligence applications. The enactment of the EU General Data Protection Regulation (GDPR) raises concerns about individuals' privacy in big data. Federated learning (FL) emerges as a functional solution that can help build high-performance models shared among multiple parties while still complying with user privacy and data confidentiality requirements. Although FL has been intensively studied and used in real applications, there is still limited research related to its prospects and applications as a FLaaS (Federated Learning as a Service) to interested 3rd parties. In this paper, we present a FLaaS system: DID-eFed, where FL is facilitated by decentralized identities (DID) and a smart contract. DID enables a more flexible and credible decentralized access management in our system, while the smart contract offers a frictionless and less error-prone process. We describe particularly the scenario where our DID-eFed enables the FLaaS among hospitals and research institutions.
BibTeX
@inproceedings{geng2021did,
  author = {Jiahui Geng and Neel Kanwal and Martin Gilje Jaatun and Chunming Rong},
  title = {{DID-eFed: Facilitating Federated Learning as a Service with Decentralized Identities}},
  booktitle = {EASE},
  year = {2021},
  url = {https://openreview.net/forum?id=6IBJJTMKkO},
}
Figure