Federated Learning in a Resource Constrained Networked Environment
Military / Coalition Issue
The ability to re-train or enhance the training of machine learning models while already deployed in order to adapt to local situations will be a key requirement. Being able to do so across a distributed set of compute capability would enhance robustness and potentially speed up learning convergence.
Core Idea and Key Achievements
This demonstration shows how the training of machine learning models can be allocated and distributed across an array of loosely networked nodes, and then recombined into a working model. In a highly distributed, low power environment, there may never be a single device with either enough processing power, or network connected for sufficient time, or with sufficient bandwidth in order to process the training set to completion in a single period of time or connectivity. This initial demonstration was based on work completed for the DAIS AFM 2019, and the underlying science has been continued and extended as part of recent work Adaptive Federated Learning in Resource Constrained Edge Computing Systems.
Implications for Defence
This solution potentially allows machine learning model retraining on devices while in the field, in order to achieve better accuracy in local conditions. By being distributed this solution reduces training time by sharing the computational burden with other local devices, while increasing resilience by not having to rely on them.
Readiness and Alternative Defence Uses
This technology while most applicable to defence, can also be used in any situation where you may have a locally distributed compute capability across a network of smaller devices, but that the communications links back to a base may be insufficient to allow reasonable data transfer of ML models.
This work has been continued by combining it with some of the coreset modelling work and model pruning. Both of these help reduce the networking requirements between devices and so increase the utility of this solution at the edge of networks.
Resources and References
- Conway-Jones, Dave, et al. “Demonstration of federated learning in a resource-constrained networked environment.” 2019 IEEE International Conference on Smart Computing (SMARTCOMP). IEEE, 2019.
IBM US, Imperial, PSU, IBM UK