Browsing by Author "Abbasi, Ali"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Open Access Carbon-aware Federated Learning with Model Size Adaptation(2024-07-23) Abbasi, Ali; Drew, Steve; Wang, Xin; Far, Behrouz; Moshirpour, MohammadDeveloping machine learning models heavily depends on the availability of data. Establishing a responsible data economy and safeguarding data ownership are essential to facilitate learning from distinct, heterogeneous data sources without centralizing data. Federated learning (FL) provides a collaborative framework that enables model development using data from geographically distributed clients, each characterized by unique carbon footprints associated with varying energy sources which can lead to significant carbon emissions when learning from these decentralized data from edge clients like smart phones and IoT devices. This variability in carbon intensity poses a substantial challenge in striving for environmentally sustainable model training with minimal carbon emissions. This thesis introduces innovative carbon-aware strategies within FL to mitigate total carbon emissions through strategic client engagement and resource allocation. We propose two distinct methods: (1) clustering clients based on data distribution and offsetting high carbon emissions with those exhibiting lower emissions, implemented through a client similarity matrix (FedGreenCS), and (2) adapting model sizes based on the carbon intensity of client locations (FedGreen), employing model compression techniques. Our results affirm the effectiveness of both approaches in harmonizing model performance with environmental impact, underscoring their potential as sustainable solutions in distributed learning scenarios. We conduct a theoretical analysis of the trade-offs between carbon emissions and convergence accuracy, taking into account the carbon intensity disparities across different regions to optimally select parameters. Empirical studies reveal that model size adaptation significantly reduces the carbon footprints of FL, surpassing contemporary methods while maintaining competitive accuracy. This research also highlights the viability of client selection and model adaptation as sustainable strategies in distributed learning contexts.Item Open Access Resource Management in Virtual Wireless Networks(2016-02-02) Abbasi, Ali; Ghaderi, Majid; Behjat, Laleh; Williamson, Carey; Fapojuwo, Abraham; Liu, JiangchuanThe main mechanism to cope with the increasing traffic demand in cellular networks is to deploy base stations more densely. A consequence of such a deployment is the higher operational expenditure imposed on mobile network operators. This happens while the revenue per bit transferred is decreasing at a fast rate which motivates designing new methods to curb the operational expenditure. A promising approach to achieve this goal is virtualization. Virtualization refers to decoupling of the physical infrastructure from its services, pooling the infrastructure’s resources, and sharing them among multiple mobile virtual network operators which can lead to better utilization of the infrastructure and lower operational expenditure for each operator. In such a system, two problems arise: 1. How to minimize the infrastructure provider’s operational cost? While there are several places to look for saving opportunities, we focus on minimizing the infrastructure’s energy consumption via dynamic activation of base stations. 2. How to minimize a mobile virtual network operator’s operational cost? The operational cost of a virtual network operator is largely dependent on the cost of acquiring bandwidth from the infrastructure provider. We focus on minimizing this cost where reservation-based and on-demand bandwidth acquisition modes are provided. The goal of this thesis is to answer these two problems. With regard to the former problem, based on the Lyapunov optimization framework, a controller is designed that dynamically adapts the network capacity to the traffic demand. Algorithms for implementing the controller in centralized and distributed settings are presented. The latter problem is formulated as a robust optimization problem. The optimal bandwidth reservation policies are derived when the low order statistics of the traffic demand are known to the mobile virtual network operator.