FLaaS: The Future of AI Training in 6G Networks

The ENABLE-6G research team has made significant progress since the project began in January of this year. The partners, IMDEA Networks, Telefónica Research, NEC Labs Europe and BluSpecs, recently met to discuss the team’s progress, goals, and challenges so far. One important area of research in the MAP-6G team focuses on Federated Learning, which involves the training of AI models in a privacy-preserving manner.

But What is Federated Learning?

AI applications such as chatbots like ChatGPT or image generators like DALL-E are extremely data-hungry and require immense amounts of data to train them in certain tasks. They are provided with millions of data points to create an AI model to complete specific tasks. Usually, when training these AIs through machine learning, the data is gathered and brought together in one place like a central server, where the training of the model happens.

With the growth of IoT, and amidst security and privacy concerns of cloud computing, many organisations are moving towards storing their data on the edge, which is where Federated Learning comes in.

Federated Learning (FL) has changed how we train these models in a way to ensure data and privacy protection of this data. Introduced by Google AI researchers in 2016, the concept of Federated Learning decentralises the training system by allowing each device to hold its own local version of the model, which sends updates back to the centralised system, not all of the raw data. The central server aggregates these updates to create what is known as the global market. Seeing as all the data is not stored in one location but spread across every user’s device, this safeguards sensitive user data for model training.

Why are We Researching It?

Federated Learning as a Service (FLaaS) has been one of the main research topics of the early research into 6G networks globally. Although 6G isn’t expected to launch until at least 2030, researchers are already investigating how to improve capabilities with the emergence of new technologies such as AI, ML and IoT. Incorporating these technologies into the next generation of wireless networks will allow for the possibility of solutions like augmented reality (AR), virtual reality (VR), telematics, autonomous vehicles, brain-computer interfaces and smart cities to operate over 6G networks.

Since its conception, many companies have been using Federated Learning to optimise and secure AI training models, with many companies providing Federated Learning as a Service (FLaaS). Designed to give users more privacy, it can allow advertisers to better deliver targeted ads to users. Google’s solution ‘Federated Learning of Cohorts (FLoC)’ is just one of many in the industry. The global FLaaS market has grown rapidly with its value expected to reach nearly $210M by 2028, with many other companies exploring the market such as IBM, Microsoft and Intel.

The Future of AI is Federated

Within the ENABLE 6G project, FLaaS is currently being studied by IMDEA Networks and Telefónica under MAP-6G. Nicolas Kourtellis, Co-Director of Telefónica Research and the Principal Scientist in MAP-6G is looking in particular at FLaaS, machine learning (ML) algorithms and privacy-preserving methods to ensure that user data is protected in the development of the network.

FLaaS offers several important benefits for 6G wireless networks. First, it helps save energy and resources by sending small model updates instead of large amounts of data to a central server. This helps conserve energy and reduces the network demand. Additionally, FLaaS reduces delays by allowing local machine learning model training on devices, which means updates happen faster and there is less time spent waiting for transmission. It also protects user data privacy by keeping data on their devices and only sharing model updates, which strengthens data security. Lastly, FLaaS improves learning performance by using Federated Learning to train multiple classifiers using different edge datasets, which can lead to better results through diverse learning experiences.

However, federated learning does not come without its challenges. It isn’t a risk-free concept although it has the potential to secure personal data more effectively. There is the risk of inference attacks where adversaries may try to extract information about the data.

Some common attacks include:

• Data / Model Poisoning
• Leak attacks
• Reverse engineering data attacks
• Poison ML Sponge Attacks on Smartphones

The ENABLE-6G team are exploring ways to reduce the gravity of these attacks and is researching ways to further protect users from these types of attacks.

A game-changer for AI

Looking ahead, FLaaS holds immense potential for the future of 6G networks. Incorporating technologies such as AI, machine learning, and IoT, will enable innovation in augmented reality, virtual reality, autonomous vehicles, and smart cities. It will also play a crucial role in optimising and securing the training of AI models to support these new applications. FLaaS is a game-changer in the world of AI and machine learning. It transforms how AI models are trained while safeguarding our sensitive data. As the ENABLE 6G project dives deeper into FLaaS, exciting inventions are on the horizon. 

Acronyms

Sources

https://www.v7labs.com/blog/federated-learning-guide#:~:text=Federated%20learning%20

https://affise.com/blog/federated-learning-of-cohorts-what-is-coming-for-the-mobile-attribution-new-era/#:~:text=Federated

https://www.marketsandmarkets.com/Market-Reports/federated-learning-solutions-market-151896843.html