The problem
In real deployments some phones drop out or are slow, delaying every round, and an attacker can join with manipulated updates that plant a backdoor. Defences against each problem exist separately; on resource-limited edge devices they interact.
Who needs it
Any company deploying federated learning on customer devices (keyboards, health, IoT).
Idea of the approach
- Simulate a federation with heterogeneous device speeds and a fraction of attackers (label flipping, backdoor).
- Implement robust aggregation (Krum, trimmed mean) and straggler handling (deadline per round, FedProx).
- Propose a combined client-selection score using speed and update trustworthiness.
- Measure accuracy, attack success rate and round time.
What you will build and deliver
- Simulation framework and defences.
- Paper.
Skills needed
Python, PyTorch, basic security notions.
Research contribution & where to publish
Contribution: joint straggler- and Byzantine-aware client selection. Venues: IEEE INFOCOM workshops, ESORICS workshops, Computers & Security, IEEE TDSC.
Key references — start reading here
- P. Blanchard, E. M. El Mhamdi, R. Guerraoui, J. Stainer. Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent. NeurIPS, 2017.
- E. Bagdasaryan, A. Veit, Y. Hua, D. Estrin, V. Shmatikov. How To Backdoor Federated Learning. AISTATS, 2020.
- T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, V. Smith. Federated Optimization in Heterogeneous Networks. MLSys, 2020.
- P. Kairouz, H. B. McMahan, et al.. Advances and Open Problems in Federated Learning. Foundations and Trends in Machine Learning, 2021.
Interested?
Send the topic title and the courses you have taken.
Contact Dr. Hamza Djigal