Policy Generation for Edge Devices in Coalitions
Military / Coalition Issue
Different parts of a coalition are governed by their own sets of policies defined as directives used to guide their actions. The vision of a distributed coalition intelligence requires a dynamic, secure and resilient information infrastructure that needs to conform to the policies of each coalition member. The appropriate policy based management framework will help to attain key attributes such as autonomous operation, composing systems together, and controlling interaction among elements.
Core idea and key achievements
Policy technologies have been used successfully in management of IT systems and networks, but prevalent approaches tend to rely on rule-based systems that rely on centralized services. Coalition environments are highly dynamic, distributed, and heterogeneous, frequently without access to a centralized infrastructure. The key achievements addressed these gaps to provide a machine learning based policy learning system demonstrated by edge devices learning how to behave when presented with a new context, situation or environment. The system developed utilises a grammar and as such does not suffer with the explainability problems of many black box machine learning systems. This is because the grammar can be inspected and understood by human users in order to fully articulate the generative policy model under which the device is currently operating.
Implications for Defence
The demonstration shows autonomous edge devices being introduced to a context with which they are not familiar. The edge devices must learn how to behave, and they can do this based on the behaviour of other devices in their local area. They are able to do this in isolation on the edge device, without connection to back end infrastructure and thus being compatible with slow and unreliable communication links. Whilst the demonstration is shown for edge devices, the technology can easily be applied to any policy learning scenario either at the edge or within the back office coalition military systems.
Readiness & alternative Defence uses
The main technology utilised is an inductive machine learning technique developed during the DAIS programme, known as FastLAS. This has been released as open source software on GitHub at https://github.com/spike-imperial/FastLAS. It has been used throughout the DAIS demonstrations and experiments in this field and is still under active development. The key functionality is in place and would require hardening for operational military usage alongside development of new code into a wider policy learning system.
Resources and references
- White, Graham, Daniel Cunnington, Mark Law, Elisa Bertino, Geeth De Mel, and Alessandra Russo. “A Comparison Between Statistical and Symbolic Learning Approaches for Generative Policy Models.” In 2019 18th IEEE International Conference On Machine Learning And Applications (ICMLA), pp. 1314-1321. IEEE, 2019.
- Cunnington, Daniel, Graham White, Mark Law, and Geeth de Mel. “A demonstration of generative policy models in coalition environments.” In International Conference on Practical Applications of Agents and Multi-Agent Systems, pp. 242-245. Springer, Cham, 2019.
- Aspis, Yaniv, Daniel Cunnington, Mark Law, Alessandra Russo, Krysia Broda, Jorge Lobo, Ankush Singla, Elisa Bertino, and Dinesh Verma. “Continuous Federated Learning of Global Policies in Coalition Environments.”
- White, Graham, John Ingham, Mark Law, and Alessandra Russo. “Using an asg based generative policy to model human rules.” In 2019 IEEE International Conference on Smart Computing (SMARTCOMP), pp. 99-103. IEEE, 2019.
- Law, Mark, Alessandra Russo, Elisa Bertino, Krysia Broda, and Jorge Lobo. “FastLAS: scalable inductive logic programming incorporating domain-specific optimisation criteria.” In Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 03, pp. 2877-2885. 2020.
- Vilamala, Marc Roig, Mark Law, Harrison Taylor, Tianwei Xing, Luis Garcia, Dave Braines, Dan Cunnington et al. “Towards Maintaining and Reusing Complex Event Processing Systems.”
- Cunnington, Daniel, Irene Manotas, Mark Law, Geeth de Mel, Seraphin Calo, Elisa Bertino, and Alessandra Russo. “A generative policy model for connected and autonomous vehicles.” In 2019 IEEE Intelligent Transportation Systems Conference (ITSC), pp. 1558-1565. IEEE, 2019.
IBM UK, Imperial, Purdue