Postdoc position in Federated Learning for Enhancing Security and Privacy of Decentralised and Distributed Systems -
Contact:
- Kandaraj Piamrat: kandaraj.piamrat@ls2n.fr
Duration: 12 months
Starting on : TBD
Context
Decentralized and distributed systems (DDS) can support businesses with needs related to computing and processing capacity in real-time. However, these systems are constantly facing challenges from the sophisticated cyber-attacks, which evolved and propagated in different parts of the system and lead to potential disruption [1]. The federated learning (FL)-based Intrusion Detection Systems (IDS) that employ synchronous techniques, where it is necessary to wait for all the FL clients to convey the global model, tend to result in prolonged duration for a single round, leading to inefficiencies [2]. Asynchronous techniques, while capable of parallel execution, often suffer from staleness due to the use of outdated models [3]. To handle these problems, semi-asynchronous techniques that combine the best of both is a promising solution. Unfortunately, it still yields suboptimal training performance. Therefore, it is crucial to develop advanced IDS that can robustly handle the complexities and challenges presented by DDS [4]. The objective is to enhance security measures, improve real-time intrusion detection, and ensure the privacy of data across all levels of decentralized and distributed systems.
This position is funded by a European project Di4SPDS (Distributed Intelligence for Enhancing Security and Privacy of Decentralised and Distributed Systems), which means that the postdoc will participate in deliverables and meetings, organized by the partners in France, Finland, Spain, and Turkey.
Subject
The main objective of this postdoc is to develop a novel collaborative IDS by integrating federated learning based on a semi-asynchronous mechanism. The proposed methods will first find the optimal number of workers that will potentially save the training time in the communication round; and second, aim to improve overall accuracy with a high intrusion detection rate and low false alarm rate. Also, we will study how the proposed methods can be applied efficiently with non-independent and identically distributed (non-IID) data as well as straggler devices with limited and heterogeneous system capacity.
The postdoc will start by analyzing the state of the art in FL/semi-asynchronous FL and pointing out the limitations of the existing methods. Novel methods will be proposed, developed, and evaluated with appropriate benchmarks then published.
Expected skills
The following skills are expected from the successful candidate:
- PhD with experience in machine learning and deep learning as well as networking and distributed system, knowledge in blockchain will be a plus
- Solid programming skills in Python language, familiar with typical deep learning frameworks (TensorFlow/PyTorch) and models.
- Good English communication skills and teamwork abilities
References
[1] Y. Xiao, Y. Jia, C. Liu, X. Cheng, J. Yu, and W. Lv, “Edge computing security: State of the art and challenges,” Proceedings of the IEEE, vol. 107, no. 8, pp. 1608-1631, 2019. [2] O. Aouedi, K. Piamrat, G. Muller, and K. Singh, “Federated semisupervised learning for attack detection in industrial internet of things,” IEEE Transactions on Industrial Informatics, vol. 19, no. 1, pp. 286-295, 2022. [3] N. Yang, D. Yuan, Y. Zhang, Y. Deng, and W. Bao, “Asynchronous semi-supervised federated learning with provable convergence in edge computing,” IEEE Network, vol. 36, no. 5, pp. 136-143, 2022. [4] W. Wu, L. He, W. Lin, R. Mao, C. Maple, and S. Jarvis, “Safa: A semi-asynchronous protocol for fast federated learning with low overhead,” IEEE Transactions on Computers, vol. 70, no. 5, pp. 655-668, 2020.
Team: Stack