On reducing the communication cost of the diffusion LMS algorithm

Ibrahim El Khalil Harrane (1) , Rémi Flamary (1) , Cédric Richard (1)

1 LAGRANGE - Joseph Louis LAGRANGE

—The recent breakthroughs in the fields of computer sciences and engineering slowly stir the world towards a more connected environment. This consequence leads to an overgrowing amount of collected data flowing from different types of devices. While it is still possible to process the incoming informations in a centralized manner, it is often more suitable to consider a distributed solution. In fact, a distributed approaches such as diffusion strategy offers more flexibility and can easily handle large amounts of data by distributing the tasks over different agents. Although it is relatively simple to implement on a cluster of data centers, diffusion strategies can be very challenging to implement on an ad-hoc based networks. Indeed, due to their limited energy budgets, ad-hoc can not sustain high communication loads. In this paper, as a first step towards the implementation of diffusion LMS on limited energy devices, we first introduce a diffusion LMS strategy that significantly reduces communication loads without compromising performance. Then, we perform analyses in the mean and mean-square sense of the proposed algorithm. Additionally, we conduct numerical experimentation to confirm the theoretical findings. Finally, we perform large scale simulations to test the algorithm efficiency in a scenario where energy is limited.

Domaines: Traitement du signal et de l'image [eess.SP]; Machine Learning [stat.ML]

Ibrahim El Khalil Harrane, Rémi Flamary, Cédric Richard. On reducing the communication cost of the diffusion LMS algorithm. IEEE Transactions on Signal and Information Processing over Networks, 2019, 5 (1), pp.100-112. ⟨10.1109/TSIPN.2018.2863218⟩. ⟨hal-01640064⟩