ContactResearch & open source
The publications below are our method evidence, not company deployment results.
Data differs between machines and sites. Clustered federated pre-training investigates how related participants can contribute to anomalous sound detection.
Federated anomaly detectionInvestigate how to weigh each participant’s contribution when local data differs, rather than assume every update is equally useful.
Adaptive learning researchSemi-supervised aggregation and unsupervised representation learning inform work with scarce labels. Self-supervised representation learning remains an ongoing direction.
Methods & evidenceContinual and lifelong adaptation remain ongoing research: address drift without forgetting earlier behavior.
Compatibility & local evaluationSelected publications
René Glitza, Luca Becker, Rainer Martin
Investigates cooperative multi-agent reinforcement learning for adaptive aggregation in semi-supervised federated learning with heterogeneous data.
Read publication: Cooperative Multi-Agent Reinforcement Learning for Adaptive Aggregation in Semi-Supervised Federated Learning with non-IID DatapFedMARL reproduction codeRené Glitza, Luca Becker, Rainer Martin
Investigates loss-weighted clustered federated pre-training for unsupervised anomalous sound detection.
Read publication: Unsupervised Anomalous Sound Detection Using Loss-Weighted Clustered Federated Pre-TrainingLuca Becker, Alexandru Nelus, René Glitza, Rainer Martin
Combines federated learning and a variational autoencoder to accelerate unsupervised clustering in acoustic sensor networks.
Read publication: Accelerated Unsupervised Clustering in Acoustic Sensor Networks Using Federated Learning and a Variational AutoencoderSoftware & reproduction