THALES SIX GTS France is leading the Cyber Secure Federated Deep Reinforcement Learning and contributing to cloud native communications security. Within NEMO, Thales will strengthen its expertise and know-how in advanced cyberattack detection and mitigation using Federated Learning, Reinforcement learning, as well as Cloud native communications, while enhancing its Advanced Studies Labs capacities. NEMO is in line with the THALES product and services roadmap and will support the TSG Advanced studies labs, in charge of de-risking novel technologies for Thales inner Product Lines, through advanced prototyping and demonstrations. 

The CFDRL component in NEMO provides capacity of learning decision-making models in a collaborative and distributed way between communicating nodes/entities. It comes with features that strengthens the cyber-security of the learning process by detecting external attacks and deploying appropriate countermeasures. The learning procedure combines two complementary learning paradigms: Federated Learning (FL) and Reinforcement Learning (RL).  

Figure 1. Federated Learning

Within the NEMO project, there are two types of applications that are foreseen:

  • Living Labs: The living labs of NEMO propose decision-making problem that could be formalized in a RL formalism. Moreover, some of them would fit the Federated Learning framework and thus fill their need for collaborative and privacy preserving learning.
  • Meta-Orchestrator:  Part of NEMO is to implement an intelligent open source meta-Orchestrator able to explore the decentralization and distribution of computing workflow over the IoT to edge to cloud continuum. NEMO meta-Orchestrator will utilise intelligence based on CF-DRL to assess strategies for orchestration of different kind of containerized Micro-services or unikernels workflows enabling migration, placement and scaling. Decision will be mainly based on three parameters: a) migration time, b) downtime, and c) overhead time, while additional functional parameters, such as network and resource availability, non-functional requirements, such as policies, energy efficiency, CO2 footprint and FinOps requirements, like networking and hosting cost will also be analysed and evaluated.