Prof. Jon Crowcroft
Title: “Some Challenges of Scaling Federated Learning”
- Abstract: In this talk, I’ll discuss some of the challenges of federated and distributed machine learning.
In the context of SmartNets, many applications have a need for local procesing and sensor input (and model based control) can be run without transmission to central servers. For some classes of application, energy and network capacity may also be limited. However, many applications also need to verify the overall system behaviour and may also use data for prediction and planning (or even billing). However it is often not necessary to send raw data, and simply sharing and aggregating locally learned model parameters may be sufficient for many use cases.
This also can help address privacy/confidentiality requirements, as well as resource contraints.
However, when building systems at scale (e.g. all UK household utility use, would have 35M sources), central model aggregation will not work, data may be drawn from multiple different distributions, including adversarial input. As we gain more experience in this space, especially with the flourishing of national and larger scale digital twins, the techniques to scale FL are moving into the mainstream needs. Luckily for us, we can often re-purpose many lessons from distributed systems of the last few decades.
- Bio: Jon Crowcroft has been the Marconi Professor of Communications Systems in the Computer Laboratory since October 2001. He has worked in the area of Internet support for multimedia communications for over 30 years. Three main topics of interest have been scalable multicast routing, practical approaches to traffic management, and the design of deployable end-to-end protocols. Current active research areas are Opportunistic Communications, Social Networks, Privacy Preserving Analytics, and techniques and algorithms to scale infrastructure-free mobile systems. He leans towards a “build and learn” paradigm for research.
From 2016-2018, he was Program Chair at the Turing, the UK’s national Data Science and AI Institute, and is now researcher-at-large there.
He graduated in Physics from Trinity College, University of Cambridge in 1979, gained an MSc in Computing in 1981 and PhD in 1993, both from UCL. He is a Fellow the Royal Society, a Fellow of the ACM, a Fellow of the British Computer Society, a Fellow of the IET and the Royal Academy of Engineering and a Fellow of the IEEE.
He likes teaching, and has published a few books based on learning materials.