[Paper Review] Federated Learning: Organizational Opportunities, Challenges, and Adoption Strategies

Joaquín Delgado Fernández, Martin Brennecke+3 more|arXiv (Cornell University)|Aug 4, 2023

Blockchain Technology Applications and SecurityComputer Science3 citations

TL;DR

This paper proposes a conceptual framework for federated learning (FL) adoption across organizations, enabling collaborative AI model training without sharing sensitive data. By mapping organizations along AI capability and data-sharing limit dimensions, it identifies tailored FL strategies for sectors like finance, healthcare, and public administration, while highlighting key technical, organizational, and regulatory research opportunities.

ABSTRACT

Restrictive rules for data sharing in many industries have led to the development of federated learning. Federated learning is a machine-learning technique that allows distributed clients to train models collaboratively without the need to share their respective training data with others. In this paper, we first explore the technical foundations of federated learning and its organizational opportunities. Second, we present a conceptual framework for the adoption of federated learning, mapping four types of organizations by their artificial intelligence capabilities and limits to data sharing. We then discuss why exemplary organizations in different contexts - including public authorities, financial service providers, manufacturing companies, as well as research and development consortia - might consider different approaches to federated learning. To conclude, we argue that federated learning presents organizational challenges with ample interdisciplinary opportunities for information systems researchers.

Motivation & Objective

• To address restrictive data-sharing regulations that hinder cross-organizational AI collaboration in industries like finance, healthcare, and public administration.
• To analyze how federated learning (FL) enables collaborative model training while preserving data privacy and competitive advantage.
• To develop a conceptual framework for FL adoption based on organizational AI capabilities and data-sharing constraints.
• To identify interdisciplinary research opportunities in technical, organizational, and regulatory dimensions of FL implementation.
• To support sustainable and equitable AI innovation through decentralized, privacy-preserving machine learning.

Proposed method

• Proposes a four-quadrant organizational framework based on AI capability and data-sharing limits to classify organizations' FL adoption potential.
• Compares centralized machine learning with federated learning (FL) architectures, emphasizing data decentralization and local model training.
• Outlines the FL training process: clients download a global model, train locally on their data, upload updated models, and aggregate them on a central server (e.g., Fed-Avg).
• Introduces key FL components such as secure aggregation (SecAgg), differential privacy (DP), and secure multiparty computation (SMPC) to enhance privacy and security.
• Applies the Technology–Organization–Environment (TOE) framework to analyze FL adoption contexts and challenges.
• Uses design science research (DSR) methodology to develop and validate the FL adoption framework through case study and grounded theory insights.

Figure 2: Federated learning at the intersection of regulation, competition, and AI capabilities.

Experimental results

Research questions

• RQ1 What are the best practices to ensure security and privacy in federated learning settings?
• RQ2 Which machine learning model architectures are most suitable for federated learning in heterogeneous, distributed data environments?
• RQ3 How can federated learning algorithms be optimized for performance across diverse organizational data sources?
• RQ4 What governance structures are required to ensure fair value distribution and equitable participation in FL collaborations?
• RQ5 How can legal and regulatory frameworks balance data privacy, innovation, and antitrust concerns in cross-jurisdictional FL deployments?

Key findings

• Federated learning enables organizations to co-create AI value without sharing raw data, preserving privacy and competitive advantage.
• The proposed four-quadrant framework classifies organizations based on AI capability and data-sharing limits, guiding tailored FL adoption strategies.
• FL reduces reliance on centralized data repositories, mitigating risks associated with data breaches and regulatory non-compliance (e.g., GDPR, CCPA).
• Secure aggregation, differential privacy, and SMPC enhance privacy and security in FL, though technical and operational challenges remain.
• Effective governance is critical to prevent conflicts of interest and ensure fair value distribution among participating organizations.
• Regulatory uncertainty persists, particularly around antitrust implications and cross-border FL deployments, necessitating clearer legal frameworks.

Figure 3: A conceptual framework for the adoption of FL