6.3.5 Secret Sharing-Based FTL
6.4 Challenges and Outlook
7 Incentive Mechanism Design for Federated Learning
7.1 Paying for Contributions
7.1.1 Profit-Sharing Games
7.1.2 Reverse Auctions
7.2 A Fairness-Aware Profit Sharing Framework
7.2.1 Modeling Contribution
7.2.2 Modeling Cost
7.2.3 Modeling Regret
7.2.4 Modeling Temporal Regret
7.2.5 The Policy Orchestrator
7.2.6 Computing Payoff Weightage
7.3 Discussions
8 Federated Learning for Vision, Language, and Recommendation
8.1 Federated Learning for Computer Vision
8.1.1 Federated CV
8.1.2 Related Works
8.1.3 Challenges and Outlook
8.2 Federated Learning for NLP
8.2.1 Federated NLP
8.2.2 Related Works
8.2.3 Challenges and Outlook
8.3 Federated Learning for Recommendation Systems
8.3.1 Recommendation Model
8.3.2 Federated Recommendation System
8.3.3 Related Works
8.3.4 Challenges and Outlook
9 Federated Reinforcement Learning
9.1 Introduction to Reinforcement Learning
9.1.1 Policy
9.1.2 Reward
9.1.3 Value Function
9.1.4 Model of the Environment
9.1.5 RL Background Example
9.2 Reinforcement Learning Algorithms
9.3 Distributed Reinforcement Learning
9.3.1 Asynchronous Distributed Reinforcement Learning
9.3.2 Synchronous Distributed Reinforcement Learning
9.4 Federated Reinforcement Learning
9.4.1 Background and Categorization
9.5 Challenges and Outlook
10.1 Finance
10.2 Healthcare
10.3 Education
10.4 Urban Computing and Smart City
10.5 Edge Computing and Internet of Things
10.6 Blockchain
10.7 5G Mobile Networks
A Legal Development on Data Protection
A.1 Data Protection in the European Union
A.1.1 The Terminology of GDPR
A.1.2 Highlights of GDPR
A.1.3 Impact of GDPR
A.2 Data Protection in the USA
A.3 Data Protection in China
Preface
This book is about how to build and use machine learning (ML) models in artificial intelligence (AI) applications when the data are scattered across different sites, owned by different individuals or organizations, and there is no easy solution to bring the data together. Nowadays, we often hear that we are in the era of big data, and big data is an important ingredient that fuels AI advances in today’s society. However, the truth is that we are in an era of small, isolated, and fragmented data silos. Data are collected and located at edge devices such as mobile phones. Organizations such as hospitals often have limited views on users’ data due to their specialties. However, privacy and security requirements make it increasingly infeasible to merge the data at different organizations in a simple way. In such a context, federated machine learning (or federated learning, in short) emerges as a functional solution that can help build high-performance models shared among multiple parties while still complying with requirements for user privacy and data confidentiality.
Besides privacy and security concerns, another strong motivation for federated learning is to maximally use the computing power at the edge devices of a cloud system, where the communication is most efficient when only the computed results, rather than raw data, are transmitted between devices and servers. For example, autonomous cars can handle most computation locally and exchange the required results with the cloud at intervals. Satellites can finish most of the computation for information that they are to gather and communicate with the earth-based computers using minimal communication channels. Federated learning allows synchronization of computation between multiple devices and computing servers by exchanging only computed results.
We can explain federated learning with an analogy. That is, an ML model is like a sheep and the data is the grass. A traditional way to rear sheep
