DPMM-CFL: Clustered Federated Learning via Dirichlet Process Mixture Model Nonparametric Clustering
Mariona Jaramillo-Civill, Peng Wu, Pau Closas.
Under review, submitted to IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2026.
[arXiv]

Bayesian Jammer Localization using Log-linear Mixture of Experts with CNN and Pathloss Models
Mariona Jaramillo-Civill, Luis González-Gudiño, Tales Imbiriba, Pau Closas.
Under review, submitted to IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2026.

Active Jammer Localization via Acquisition-Aware Path Planning
Luis González-Gudiño, Mariona Jaramillo-Civill, Pau Closas, Tales Imbiriba.
Under review, submitted to IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2026.

Trends and Challenges in Next-Generation GNSS Interference Management
Leatile Marata, Mariona Jaramillo-Civill, Tales Imbiriba, Petri Välisuo, Heidi Kuusniemi, Elena Simona Lohan, Pau Closas.
Under review, submitted to IEEE Aerospace and Electronic Systems Magazine, 2025.

Jammer Source Localization with Federated Learning
Mariona Jaramillo-Civill, Peng Wu, Andrea Nardin, Tales Imbiriba, Pau Closas.
IEEE/ION Position, Location and Navigation Symposium (PLANS), 2025. [Oral Presentation; Best Paper nominee]
[full paper] [conference] [code]

Clustered Federated Learning (CFL) with Unknown Number of Clusters

• Addresses client heterogeneity in federated learning via Bayesian nonparametric clustering.
• Uses a Dirichlet Process Mixture Model (DPMM) to dynamically infer the number of clusters while training.
• Employs restricted Gibbs sampling with split–merge proposals and weighted aggregation.
• Achieves scalable personalization without fixing the number of clusters K in advance.

CFL DPMM Split–Merge MCMC

Bayesian, Federated, and Active Learning Approaches for Jammer Localization

• Develops ML models for privacy-preserving GNSS jammer localization in crowdsourced environments.
• APBM: Differentiable pathloss model + neural network, trained with FedAvg for decentralized optimization.
• Bayesian mixtures of models: Pathloss expert combined with CNN to capture contextual effects and quantify uncertainty.
• Active learning: Used to guide data collection toward the most informative regions.
• Improves localization accuracy and robustness in complex multipath-affected urban settings.

Predicted jammer field and source localization in an urban scenario using the Augmented Physics-Based Model (APBM).

APBM CNN FedAvg Bayesian Mixtures Active Learning