Ph.D Position on Federated learning with low resources and privacy
TinyPART will thus provide both (i) capabilities to isolate small runtime containers of untrusted (possibly scripted) IoT logic, and (ii) adequate privacy-oriented preprocessing (such as differential privacy and lightweight cryptographic tools) of IoT data on-board, before it is transferred from the OS to the container(s) and/or from the containers to the cloud. A typical example addressed in TinyPART is federated machine learning which maintains a local model of the targeted learning task while aggregating protected data centrally in the cloud. While (particularly Python-scripted) federated machine learning will guide the design, TinyPART aims to retain generic applicability in the IoT realm. The platform will be demonstrated in the context of the open source RIOT OS and PIP kernel. While contributing in the creation of isolated, flexible and updatable runtime containers for constrained devices, the project will seek the best tradeoff between isolation guarantees, the logic orchestration functionality & security, memory footprint and ease of use by non-specialist embedded systems developer.