Machine learning has demonstrated great potential to revolutionize the Internet of Things (IoT) by improving the efficiency of deployment and management of IoT, enhancing IoT security and privacy protection, and enabling various smart applications. The success of artificial intelligence (AI) in IoT stems from availability of big training data and massive computation power, which drives academia and industry to build huge ICT infrastructure for data storage, transmission, and processing. However, in many applications, training data are generated by IoT devices owned by individuals, who hesitate to share their data that expose privacy. Moreover, the amount of data generated by IoT devices will dramatically grow in the 5G/beyond-5G (B5G) era.
There are nearly 8 billion IoT devices and smartphones interconnected around the world, and that number will increase to 80 billion by 2025, as estimated by Intersectional Data Corporation (IDC). Another report from Cisco shows that nearly 847 ZB of data will be generated at the network edge by 2021. Therefore, it is not appropriate or even feasible to gather the data for training in a centralized datacenter. Federated learning has been proposed to enable distributed computing nodes to collaboratively train models without exposing their own data. Its basic idea is to let these computing nodes train local models using their own data, respectively, and then upload the local models, instead of raw data, to a logically centralized parameter server that synthesizes a global model. Since its inception by Google, federated learning has shown great promises in protecting data privacy and reducing network traffic. However, there are still many open challenges that are unsolved. It is expected that IoT devices will be empowered by 5G/B5G networks with extremely low latency and high bandwidth, so that they can join federated learning by contributing data and computation resources in a more efficient way. It is critical to optimize computation and network resources to accelerate the federated-learning process. Meanwhile, federated learning can motivate new business models of trading values via learning, instead of trading data in existing crowdsensing schemes. Therefore, it is necessary to study the federated learning for IoT using economic approaches.
Addressing the above challenges of federated learning in 5G/B5G networks needs significant research efforts on theories, algorithms, architecture, and experiences of system deployment and maintenance. Therefore, this special issue aims to offer a platform for researchers from both academia and industry to publish recent research findings and to discuss opportunities, challenges, and solutions related to federated learning. In particular, this special issue solicits original research papers about state-of-the-art approaches, methodologies, and technologies enabling efficient and practical federated learning for 5G/B5G-envisoned IoT. Topics of interests include, but are not limited to:
- Federated learning theories and algorithms for IoT
- Federated learning architecture for IoT
- Efficient networking and communication of federated learning for IoT
- Security and privacy of federated learning for IoT
- Incentive mechanisms of federated learning for IoT
- Federated learning systems for IoT
- Federated learning applications for IoT
- Energy efficiency of federated learning for IoT
Submissions Due: CLOSED
Notification to Authors: May 1, 2020
Revised Manuscript Due: May 15, 2020
Final Notification: June 1, 2020
Publication in 2020
Visit the Author Information page for details on how to submit.
Peng Li, University of Aizu, Japan (email@example.com)
Amiya Nayak, University of Ottawa, Canada (firstname.lastname@example.org)
Milos Stojmenovic, Singigunum University, Serbia (email@example.com)