Decentralized Federated Learning: Enabling Collaborative AI With Enhanced Trust and Efficiency

This tutorial presents a rigorous examination of Decentralized Federated Learning (DFL), an emergent field that fortifies the intersection of collaboration, efficiency, and privacy in artificial intelligence systems. The tutorial systematically explores the underlying principles, architectures, components, security strategies, and optimization techniques inherent to DFL. Moreover, participants will have the opportunity to gain hands-on experience with Fedstellar, a platform for training DFL models, and delve into real-world use cases, effectively marrying theory with practice. Attendees will gain an understanding of practical applications, as well as a vision of future trends, encapsulating the transformative potential of DFL in the realm of Artificial Intelligence (AI) and associated domains.

ECAI is Europe’s premier AI research venue. They are honored to host the 26th edition of this conference in person in Kraków, one of Europe’s most stunning cities. ECAI 2023 will feature, as usual, a full program of technical papers, as well as many AI-related events of interest for researchers, students, and industry professionals.

The tutorial will be held at the 26th European Conference on Artificial Intelligence (ECAI 2023), which takes place on September 30-October 5 in Kraków, Poland.

Spotlight (90-minute)

Topics to be dealt with in the tutorial:

Theory Part – Decentralized Federated Learning Motivation and Fundamentals (40 minutes)

• Introduction: The Growing Importance of Decentralized Federated Learning
  • Definition and Concept
  • Motivations and Objectives
  • Advantages and Limitations
• Fundamental aspects of Decentralized Federated Learning
  • Federation Architecture
  • Network Topology
  • Communications Mechanisms
  • Security and Privacy
  • Key Performance Indicators
  • Optimizations

Practical Part – Frameworks, Applications, and Use Cases (30 minutes)

• A quick overview of Decentralized Federated Learning Frameworks, its Key Features, and Capabilities
• Fedstellar: A Platform for Decentralized Federated Learning
• Practical Exercise
  • Simulated deployments using docker containers and well-known distributed datasets
  • Decentralized Federated learning to detect anomalies produced by malware affecting Raspberry Pis

Wrap-up – Conclusions and Challenges (20 minutes)

• Emerging Trends and Technological Developments in Decentralized Federated Learning
• Open Challenges and Future Research Opportunities
• Recap of Key Insights from Decentralized Federated Learning Research

This tutorial will benefit a broad spectrum of ECAI attendees. It is particularly suitable for AI researchers, practitioners, and specialists who seek to augment their understanding of privacy-preserving learning methods and decentralized architectures. Furthermore, it is appropriate for professionals operating at the intersection of AI and other fields such as healthcare, manufacturing, and telecommunications, where DFL has substantial real-world implications. While a basic understanding of machine learning concepts and federated learning is helpful, the tutorial is structured to be accessible to those with varying levels of familiarity, from relative newcomers to seasoned experts.

Decentralized Federated Learning (DFL): As the Artificial Intelligence field continuously evolves to address growing concerns about privacy and efficiency, DFL emerges as a significant focus. This tutorial provides an in-depth understanding of the core principles, motivations, benefits, and challenges of DFL. It elucidates the comparison of centralized and decentralized approaches and delves into key architectural considerations, vital components, and network topologies. The tutorial further explores the role of communication protocols, privacy and security measures, and the importance of trust management in the context of DFL.

Federation Topologies: This tutorial extends to cover the importance and role of various federation topologies in DFL. Different federation structures significantly impact the efficiency and resilience of DFL models. We will explore the practical implications and trade-offs of various topologies, such as star, ring, or fully connected networks, and their impact on model performance, communication overhead, and fault tolerance.

DFL Frameworks: There is a scarcity of DFL frameworks available for practical implementation. This tutorial includes the status of existing solutions, analyzing their key features, capabilities, use cases, and ease of use. This section is crucial for understanding the toolsets available for practitioners wanting to implement DFL in their respective fields.

Use Cases & Practical Exercise: The transformative potential of DFL is best demonstrated through real-world applications. This tutorial offers a comprehensive overview of practical use cases across various sectors, including healthcare, Industry 4.0, mobile services, military, and autonomous vehicles. It elucidates how DFL is applied to solve pressing challenges in these sectors, providing participants with a vivid understanding of its real-world implications. Moreover, a practical exercise is included in the tutorial to allow attendees to implement a DFL model within a simulated real-world scenario, providing them an invaluable hands-on experience.