Federated vs. Centralized Machine Learning under Privacy-elastic Users: A Comparative Analysis

Drainakis, Georgios; Κατσαρός, Κωνσταντίνος; Pantazopoulos, Panagiotis; Sourlas, Vasilis; Amditis, Angelos

doi:10.1109/nca51143.2020.9306745

Cited by 45 publications

(19 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Analytical calculations are derived to capture the bandwidth and consumed energy of each scheme along with the way the involved energy costs are broken down into the system's stakeholders; thus, revealing each scheme's benefits and emerging trade-offs. This study extends our previous work [12] on the CL-FL comparison by introducing EL as an alternative distributed ML scheme; we extend the system model to allow for edge node utilization and shed light on the resources consumption per-stakeholder. Our theoretical analysis and numerical results recommend for ML workloads an edge node deployment at a regional level, closer to the cloud.…”

Section: Introductionmentioning

confidence: 61%

See 1 more Smart Citation

On the Resource Consumption of Distributed ML

Drainakis

Pantazopoulos

Κατσαρός

et al. 2021

2021 IEEE International Symposium on Local and Metropolitan Area Networks (LANMAN)

Self Cite

View full text Add to dashboard Cite

The convergence of Machine Learning (ML) with the edge computing paradigm has paved the way for distributing processing-heavy ML tasks to the network's extremes. As the edge deployment details still remain an open issue, distributed ML schemes tend to be network-agnostic; thus, their effect on the underlying network's resource consumption is largely ignored.In our work, assuming a network tree structure of varying size and edge computing characteristics, we introduce an analytical system model based on credible real-world measurements to capture the end-to-end consumption of ML schemes. In this context, we employ an edge-based (EL) and a federated (FL) ML scheme and in-depth compare their bandwidth needs and energy footprint against a cloud-based (CL) baseline approach.Our numerical evaluation suggests that EL exhibits a minimum of 25% bandwidth-efficiency compared to CL and FL, if employed by a few nodes higher in the edge network, while halving the network's energy costs.

show abstract

Section: Introductionmentioning

confidence: 61%

“…setting. The ML task terminates when all data is depleted, since our previous work in [12] has shown that utilizing the complete dataset once (essentially one epoch) with the above-mentioned model, suffices for over 75% accuracy for FL and over 80% for CL.…”

Section: Numerical Resultsmentioning

confidence: 99%

On the Resource Consumption of Distributed ML

Drainakis

Pantazopoulos

Κατσαρός

et al. 2021

2021 IEEE International Symposium on Local and Metropolitan Area Networks (LANMAN)

Self Cite

View full text Add to dashboard Cite

show abstract

“…The simulation environment was set-up in a single desktop machine with the following characteristics: Intel Core i7-10700 CPU @ 2.9 GHz, 64-bit, RAM 16 GB, OS Windows 10. We have chosen the following scenario, based on our previous work [19];…”

Section: Simulation Environment and Resultsmentioning

confidence: 99%

A Distributed ML Framework for Service Deployment in the 5G-based Automotive Vertical

Sourlas

Rizk

Κατσαρός

et al. 2021

2021 IEEE International Mediterranean Conference on Communications and Networking (MeditCom)

Self Cite

View full text Add to dashboard Cite

5G is the convergence technology for the new generation of mobile networks, expected to be massively deployed in the coming years. Building on network slicing and edge computing capabilities, 5G promises to address the diverse and quite demanding performance requirements of a wide range of use cases (UCs). As a result of these technological transformations, vertical industries will have enhanced technical capacity available to trigger the development of new products and services. Driven by these advances, Machine learning (ML) applications are headed towards collaborative distributed (CDL) schemes, to exploit the abundance of clients' data. Contrary to the traditional cloud-based centralized solutions (CML), in CDL schemes, computational load is shifted to the intelligent edge and extends further beyond, to the user-equipment (including connected vehicles). Here, we present a distributed ML (DML) framework, that will provide functionalities for simplified management and orchestration of collections of ML service components and will allow ML-based applications to penetrate the Automotive world.

show abstract

“…Learning Architecture. ML algorithms are generally designed with a centralized or Standalone Approach (SA) [106] in which the training data is fed to a centralized machine where training and decisions are made. However, data is generally scattered and massive for network applications, whereas the centralized approach requires massive computational resources.…”

Section: 24mentioning

confidence: 99%

A Survey on Requirements of Future Intelligent Networks: Solutions and Future Research Directions

2022

View full text Add to dashboard Cite

The context of this study examines the requirements of Future Intelligent Networks (FIN), solutions, and current research directions through a survey technique. The background of this study is hinged on the applications of Machine Learning (ML) in the networking field. Through careful analysis of literature and real-world reports, we noted that ML has significantly expedited decision-making processes, enhanced intelligent automation, and helped resolve complex problems economically in different fields of life. Various researchers have also envisioned future networks incorporating intelligent functions and operations with the ML. Several efforts have been made to automate individual functions and operations in the networking domain; however, most of the existing ML models proposed in the literature lack several vital requirements. Hence, this study aims to present a comprehensive summary of the requirements of FIN and propose a taxonomy of different network functionalities that needs to be equipped with ML techniques. The core objectives of this study are to provide a taxonomy of requirements envisioned for end-to-end FIN, relevant ML techniques, and their analysis to find research gaps, open issues, and future research directions. The real benefit of machine learning applications in any domain can only be ensured if intelligent capabilities cover all its components. We observed that future generations of networks are heterogeneous, multi-vendor, and multidimensional, and ML can provide optimal results only if intelligent capabilities are used on a holistic scale. Realizing intelligence on a holistic scale is only possible if the ML algorithms can solve heterogeneous problems in a multi-vendor and multidimensional environment. ML models must be reliable and efficient, support distributed learning architecture, and possess the capability to learn and share the knowledge across the network layers and administrative domains to solve issues. Firstly, this study ascertains the requirements of the FIN and proposes their taxonomy through reviews on envisioned ideas by various researchers and articles gathered from reputed conferences and standard developing organizations using keyword queries. Secondly, we have reviewed existing studies on ML applications focusing on coverage, heterogeneity, distributed architecture, and cross-domain knowledge learning and sharing. Our study observed that in the past, ML applications were focused mainly on an individual/isolated level only, and aspects of global and deep holistic learning with cross-layer/domain knowledge sharing with agile ML operations are not explored at large. We recommend that the issues mentioned above be addressed with improved ML architecture and agile operations and propose ML pipeline-based architecture for FIN. The significant contribution of this study is the impetus for researchers to seek ML models suitable for a modular, distributed, multi-domain and multi-layer environment and provide decision-making on a global or holistic rather than individual function level.

show abstract

Federated vs. Centralized Machine Learning under Privacy-elastic Users: A Comparative Analysis

Cited by 45 publications

References 15 publications

On the Resource Consumption of Distributed ML

On the Resource Consumption of Distributed ML

A Distributed ML Framework for Service Deployment in the 5G-based Automotive Vertical

A Survey on Requirements of Future Intelligent Networks: Solutions and Future Research Directions

Contact Info

Product

Resources

About