Flaas

What is it and what is it intended for?

FLaaS is a service that enables that enables SMEs and other third-party entities (e.g., mobile application owners) to build Machine Learning (ML) models that help their apps (e.g., for better recommendations, user modelling, etc.). Crucially the ML learning is performed on user devices, and not on the servers of the entities, using the principles of Federated Learning (FL). This means the entities do not need to collect user data at the backend (as they have been doing so far), with all the privacy implications this practice entails. Instead, they can select and train ML models on the user devices in a secure and privacy-preserving fashion. The target audience of this service are entities that may not have the resources (personnel, tech, etc.) to make this effort on their own. Interestingly, FLaaS can also enable 2 or more third-party entities to collaborate with each other to build joint ML models that serve all interested parties.

Why is it important?

This service is important as it can push the ML era into the next step, which is decentralized and distributed ML. The step we have been living so far has been the ML-as-a-Service (MLaaS): companies collecting and analyzing data of users on their cloud servers and modelling their behavior. The next step is to do this using FLas- a-Service (FLaaS): companies leaving the data at their owners (users) and using privacy-preserving ML techniques, they can build models that are still useful while protecting and respecting user privacy. We expect this market to pick-up in the next few years: there are already a dozen start-ups offering some sort of FL services for cross-silo training, and also big players such as Google, Facebook and Apple using FL inside their products, services and devices.

Where can we find more information about this?

  1. Full paper (Online, under review)
  2. Demo at Proceedings of the 20th Annual International Conference on Mobile Systems, Applications and Services (MobiSys 2022).
  3. MobiSys Poster 
  4. Demo at Proceedings of the 23rd ACM Annual International Workshop on Mobile Computing Systems and Applications (HotMobile 2022).
  5. Concept paper at Proceedings of the 1st Workshop on Distributed Machine Learning (DistributedML 2020). & FLaaS: Federated Learning as a Service
  6. FLaaS – Practical Federated Learning as a Service for Mobile Applications
CONCORDIA DELIVERS