This site loads optional analytics from Google and external analytics providers only if you accept. You can decline and continue using the site normally.
Node-to-node and semi-decentralized learning systems for security-critical environments where raw data cannot be centralized.
Decentralized Federated Learning (DFL) studies how multiple organizations, devices or agents can train models collaboratively without moving raw data to a central server. This matters when centralization is impractical, legally sensitive or operationally fragile, such as in defense, IoT, industrial monitoring and cross-organization cybersecurity.
The hardest DFL scenarios combine non-IID data, heterogeneous devices, intermittent connectivity and adversarial behavior. A robust system must decide which peers to trust, how to aggregate information, how to detect malicious updates and how to continue learning when parts of the federation are unstable.
The related work in this site covers DFL frameworks, situational awareness, reputation, prototype exchange, ensemble distillation and security monitoring. The goal is not only to train distributed models, but to make their behavior understandable, resilient and useful in security-critical deployments.
Standard federated learning usually relies on a central coordinator. DFL removes or reduces that central role so peers can exchange updates, models or representations directly or through semi-decentralized structures.
Cyberdefense often involves sensitive telemetry distributed across organizations or devices. DFL can support collaborative detection and mitigation while keeping raw evidence local.
The main barriers are unreliable peers, non-IID data, communication limits, adversarial clients, privacy risk in shared updates and the lack of mature operational tooling.
DEFENDIS develops a framework for uniquely identifying IoT devices in a distributed manner while solving security threats through decentralized federated learning.
View ProjectA project with armasuisse Cyber-Defence Campus investigating trustworthy threat intelligence sharing and verification algorithms.
View ProjectA completed collaboration with armasuisse Cyber-Defence Campus that developed robust, resource-efficient and adaptive learning for multimodal sensor ecosystems.
View ProjectDeveloped decentralized AI solutions addressing computational constraints and trust validation under armasuisse Cyber-Defence Campus sponsorship.
View ProjectA measured research agenda for dynamic topologies, manipulated context, formal privacy analysis and external validity in robust DFL.
A design note on distributed or decentralized MoE systems, with explicit terminology, routing trade-offs and safeguards for heterogeneous edge deployments.
A practical analysis of payload size, round cost and the prototype communication results reported for Modalis on AVMNIST.
An original technical walkthrough of the Modalis prototype-based approach for decentralized learning with non-IID classes and missing modalities.
Why decentralized multimodal learning needs explicit missing-modality handling, representation alignment and evaluation beyond a global average.