Fair and Robust Federated Learning

Research on fairness and robustness in federated learning for distributed systems

In settings characterized by edge nodes that collect heterogeneous and complex data, distributed training offers an effective solution for building collective machine learning models while safeguarding the privacy of local information. My research addresses some fundamental challenges of federated training, operating in non-ideal scenarios where participants have limited computational resources and may introduce noise, inaccurate labels, or exhibit malicious, uncooperative, or evasive behavior.

Selfishness and fairness in federated learning

A core focus of my research involves protecting global FL models from dishonest or non-cooperative participants. In decentralized deployments, “selfish” clients (Augello et al., 2024) may intentionally manipulate local training updates to skew the global model toward their specific local data distribution. This behavior lead to biased global models that fail to generalize across the entire population, resulting in unfair treatment of other participants and potentially degrading overall model performance. To counter this, while recovering useful information from selfish clients, I have worked on robust aggregation strategies to recover the true updates of selfish participants, while also ensuring fairness in the global model by preventing any single participant from disproportionately influencing the outcome (Augello et al., 2026).

On the other hand, if data heterogeneity causes incompatible updates, I have also explored the used of dynamic clustering of participants (Augello et al., 2023) to group clients with similar data distributions, allowing for the creation of specialized sub-models that better capture the diversity of the data while still contributing to the overall federated learning process. This approach is meant to provide each participant with a model that is more aligned with their local data distribution, quickly adapting to changes in the local data, improving both fairness and performance across the network.

Resource-aware federated learning

Battery constraints often lead to client dropouts and uneven participation. To make federated learning practical for hardware-constrained edge devices, I have worked on resource-aware frameworks (Augello et al., 2025) to coordinates battery-conscious nodes using intelligent client and data selection mechanisms, prolonging node lifespan and ensures equitable energy consumption across participating edge devices without sacrificing global model quality.

Hardware dishomogeneity can also limit the effectiveness of federated learning: either the model is too expensive for low-end devices, limiting how many clients can participate, or the model is too small to capture the complexity of the data. To address this, I have explored the use of heterogeneous model architectures (Augello et al., 2025) that allow devices with different computational capabilities to contribute to the training process, enabling a more inclusive and effective federated learning ecosystem. A similar issue arises for the feature extraction process, where some clients might not be able to extract all the features from their local data due to hardware limitations. In this context, I have explored heterogeneous graph neural networks (Augello et al., 2026) that allow clients to extract different sets of features from their local data, while still contributing to the global model training process.

References

2026

FairRFL: Fair and Robust Federated Learning in the Presence of Selfish Clients

Andrea Augello, Ashish Gupta, Giuseppe Lo Re, and 1 more author

IEEE Transactions on Emerging Topics in Computing, 2026

DOI Bib

@article{augello2026fairrfl,
  author = {Augello, Andrea and Gupta, Ashish and Lo Re, Giuseppe and Das, Sajal K.},
  journal = {IEEE Transactions on Emerging Topics in Computing},
  title = {FairRFL: Fair and Robust Federated Learning in the Presence of Selfish Clients},
  year = {2026},
  volume = {14},
  number = {1},
  issn = {2168-6750},
  pages = {316-331},
  doi = {10.1109/TETC.2026.3661199},
}

Resource-Aware Federated Learning for Malware Detection on Smart Devices

Andrea Augello, Alessandra De Paola, and Giuseppe Lo Re

In 2026 IEEE International Conference on Smart Computing (SmartComp), 2026

DOI Bib

@inproceedings{augello_2026_resourceaware,
  author = {Augello, Andrea and De Paola, Alessandra and Lo Re, Giuseppe},
  booktitle = {2026 IEEE International Conference on Smart Computing (SmartComp)},
  title = {Resource-Aware Federated Learning for Malware Detection on Smart Devices},
  year = {2026},
  volume = {},
  number = {},
  pages = {48-55},
  keywords = {Modeling;Malware;Printing;Training;Federated learning;Signal detection;Conferences;Servers;Timing;Operating systems;malware detection;federated learning;explainable AI;edge computing},
  doi = {10.1109/SmartComp69968.2026.00019},
  issn = {2693-8340 }
}

2025

BatteryFL: Battery-Aware Federated Learning

Andrea Augello, Priyesh Ranjan, Ashish Gupta, and 3 more authors

In GLOBECOM 2025 - 2025 IEEE Global Communications Conference, 2025

DOI Bib

@inproceedings{augello2025batteryfl,
  author = {Augello, Andrea and Ranjan, Priyesh and Gupta, Ashish and Corò, Federico and Lo Re, Giuseppe and Das, Sajal K.},
  booktitle = {GLOBECOM 2025 - 2025 IEEE Global Communications Conference},
  title = {BatteryFL: Battery-Aware Federated Learning},
  year = {2025},
  volume = {},
  number = {},
  pages = {1938-1943},
  isbn = {979-8-3315-7781-0},
  doi = {10.1109/GLOBECOM59602.2025.11432132}
}

Federated hyperdimensional ensembles for mobile malware detection

Andrea Augello, Alessandra De Paola, Giuseppe Lo Re, and 1 more author

In CEUR Workshop Proceedings - Joint Proceedings of the Thematic Workshops at Ital-IA 2025 colocated with the 5th National Conference on Artificial Intelligence, organized by CINI (Ital-IA 2025) , 2025

Bib

@inproceedings{augello_2025_fedhdc,
  title = {Federated hyperdimensional ensembles for mobile malware detection},
  author = {Augello, Andrea and De Paola, Alessandra and Lo Re, Giuseppe and Zangara, Gioacchino},
  booktitle = {CEUR Workshop Proceedings -  Joint Proceedings of the Thematic Workshops at Ital-IA 2025 colocated with the 5th National Conference on Artificial Intelligence, organized by CINI (Ital-IA 2025) },
  issn = {16130073},
  publisher = {CEUR-WS},
  volume = {4121},
  url = {https://ceur-ws.org/Vol-4121/Ital-IA_2025_paper_44.pdf},
  year = {2025}
}

2024

Tackling Selfish Clients in Federated Learning

Andrea Augello, Ashish Gupta, Giuseppe Lo Re, and 1 more author

In 27th European Conference on Artificial Intelligence (ECAI 2024), 2024

DOI Bib

@inproceedings{augello_2024_tackling,
  title = {Tackling Selfish Clients in Federated Learning},
  author = {Augello, Andrea and Gupta, Ashish and Lo Re, Giuseppe and Das, Sajal K},
  booktitle = {27th European Conference on Artificial Intelligence (ECAI
  		  2024)},
  pages = {1888--1895},
  year = {2024},
  doi = {10.3233/FAIA240702},
  publisher = {IOS Press},
  isbn = {978-1-64368-548-9}
}

2023

DCFL: Dynamic Clustered Federated Learning under Differential Privacy Settings

Andrea Augello, Giulio Falzone, and Giuseppe Lo Re

In 2023 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops), 2023

DOI Bib

@inproceedings{augello_2023_dcfl,
  author = {Augello, Andrea and Falzone, Giulio and Lo Re, Giuseppe},
  booktitle = {2023 IEEE International Conference on Pervasive Computing
  		  and Communications Workshops and other Affiliated Events
  		  (PerCom Workshops)},
  title = {{DCFL}: {D}ynamic {C}lustered {F}ederated {L}earning under
  		  {D}ifferential {P}rivacy {S}ettings},
  year = {2023},
  pages = {614-619},
  doi = {10.1109/PerComWorkshops56833.2023.10150285},
  isbn = {978-1-6654-5381-3}
}