Abbas Mehrabi

Pranvera Kortoçi, Yilei Liang, Pengyuan Zhou et al.

Mobile devices and the immense amount and variety of data they generate are key enablers of machine learning (ML)-based applications. Traditional ML techniques have shifted toward new paradigms such as federated (FL) and split learning (SL) to improve the protection of user's data privacy. However, these paradigms often rely on server(s) located in the edge or cloud to train computationally-heavy parts of a ML model to avoid draining the limited resource on client devices, resulting in exposing device data to such third parties. This work proposes an alternative approach to train computationally-heavy ML models in user's devices themselves, where corresponding device data resides. Specifically, we focus on GANs (generative adversarial networks) and leverage their inherent privacy-preserving attribute. We train the discriminative part of a GAN with raw data on user's devices, whereas the generative model is trained remotely (e.g., server) for which there is no need to access sensor true data. Moreover, our approach ensures that the computational load of training the discriminative model is shared among user's devices-proportional to their computation capabilities-by means of SL. We implement our proposed collaborative training scheme of a computationally-heavy GAN model in real resource-constrained devices. The results show that our system preserves data privacy, keeps a short training time, and yields same accuracy of model training in unconstrained devices (e.g., cloud). Our code can be found on https://github.com/YukariSonz/FSL-GAN

Abbas Mehrabi, Aresh Dadlani, Seungpil Moon et al.

Fluctuations in electricity tariffs induced by the sporadic nature of demand loads on power grids has initiated immense efforts to find optimal scheduling solutions for charging and discharging plug-in electric vehicles (PEVs) subject to different objective sets. In this paper, we consider vehicle-to-grid (V2G) scheduling at a geographically large scale in which PEVs have the flexibility of charging/discharging at multiple smart stations coordinated by individual aggregators. We first formulate the objective of maximizing the overall profit of both, demand and supply entities, by defining a weighting parameter. We then propose an online decentralized greedy algorithm for the formulated mixed integer non-linear programming (MINLP) problem, which incorporates efficient heuristics to practically guide each incoming vehicle to the most appropriate charging station (CS). The better performance of the presented algorithm compared to an alternative allocation strategy is demonstrated through simulations in terms of the overall achievable profit and flatness of the final electricity load. Moreover, the results of simulations reveal the existence of optimal number of deployed stations at which the overall profit can be maximized.

Abbas Mehrabi, Matti Siekkinen, Antti Ylä-Jääski

MPEG has recently proposed Server and Network Assisted Dynamic Adaptive Streaming over HTTP (SAND-DASH) for video streaming over the Internet. In contrast to the purely client-based video streaming in which each client makes its own decision to adjust its bitrate, SAND-DASH enables a group of simultaneous clients to select their bitrates in a coordinated fashion in order to improve resource utilization and quality of experience. In this paper, we study the performance of such an adaptation strategy compared to the traditional approach with large number of clients having mobile Internet access. We propose a multi-servers multi-coordinators (MSs-MCs) framework to model groups of remote clients accessing video content replicated to spatially distributed edge servers. We then formulate an optimization problem to maximize jointly the QoE of individual clients, proportional fairness in allocating the limited resources of base stations as well as balancing the utilized resources among multiple serves. We then present an efficient heuristic-based solution to the problem and perform simulations in order to explore parameter space of the scheme as well as to compare the performance to purely client-based DASH.