DATRIS (Decentralized AI for Trustworthy and Resource-efficient Intelligent Systems) is a completed research project supported by the Cyber-Defence Campus at armasuisse (Switzerland) during 2024. The research was conducted at the University of Murcia's CyberDataLab.
Project Overview
Decentralized federated systems face two critical challenges when deploying Machine Learning models to resource-constrained devices: high computational/communication overheads and vulnerability to collaborative poisoning attacks.
DATRIS addresses both challenges by studying mechanisms to optimize and secure collaborative AI. The project investigates model compression and pruning techniques alongside robust defenses such as Moving Target Defense (MTD) and LLM-assisted analysis for decentralized environments exposed to communication threats and software-level risks.
Core Accomplishments
- Resource Efficiency: Studied hardware-aware pruning and quantization to reduce model size for low-power edge gateways.
- Threat Mitigation via LLMs: Researched how Large Language Models (LLMs) may support interpretation of system calls and anomalous execution patterns in edge telemetry.
- Decentralized Protection (MTD): Explored Moving Target Defense (MTD) strategies for randomizing communication protocols and topologies, making DFL networks harder to map and exploit.
The DATRIS project was sponsored by the armasuisse Cyber-Defence Campus.
