DECIMAL (Decentralized Intelligent Multimodal Adaptative Learning for IoT Malware Detection) was a research project supported by the Cyber-Defence Campus at armasuisse (Switzerland) during 2025. It was led by the Department of Information and Communications Engineering at the University of Murcia.
Project Overview
In the Internet of Things (IoT) landscape, devices produce highly heterogeneous data modalities, including network packet headers, system logs, device temperature time-series, and process execution states. Standard Decentralized Federated Learning (DFL) models struggle to learn efficiently when nodes have access to different subsets of these modalities.
DECIMAL addressed this limitation by developing robust and resource-efficient learning models capable of handling multimodal sensor data in a fully decentralized network. The project focused on enabling IoT devices to collaboratively detect malware and security anomalies without relying on a central coordinator.
Core Pillars
- Multimodal Data Integration: Designs algorithms for learning from different combinations of modalities under data heterogeneity across directly connected nodes.
- Trust & Reputation Management: Studies decentralized trust assessment metrics to reduce the influence of poisoning attacks and malicious data injection.
- Bandwidth Optimization: Focuses on representation-level knowledge exchange, enabling nodes to share compact class-level prototypes instead of transmitting large model weights over resource-constrained networks.
The DECIMAL project is conducted in collaboration with armasuisse Cyber-Defence Campus.
