Showing papers by "Jon Crowcroft published in 2023"
TL;DR: In this article , the authors propose a data redistribution phase that balances the data distribution on different participating devices to a certain degree, which can further increase the system performance in the training phase.
Abstract: Recently academia and industry has growing interest in the sixth generation network, which aims to support a rich range of applications with higher capacity and greater coverage than existing 5G connections. One of such promising applications that can benefit from 6G is Decentralised Federated Learning, a privacy-preserving machine learning paradigm. Also, it relies heavily on peer-to-peer mobile connection among edge and mobile devices, instead of a powerful central server on the cloud. However, the data and device heterogeneity, and highly dynamic environment in mobile networks pose challenges to the performance of federated learning. In this paper, we propose a data redistribution phase that balances the data distribution on different participating devices to a certain degree, which can further increase the system performance in the training phase. To derive our method, we first model this problem as a bargaining game, the equilibrium of which is formalised as an optimisation problem. Then we propose two algorithms to solve it: a centralised one, and a decentralised one that each participant executes without centralised coordination. We further improve the energy efficiency of the decentralised algorithm by introducing several heuristics. We evaluate the proposed system with both simulation and DNN training tasks on large scale FEMNIST-based datasets.
2 citations
TL;DR: In this article , the authors established a Node and Link probabilistic failure model in the presence of node and communication link failures for a representative crash fault-tolerant distributed consensus protocol: RAFT and derived the analytical results in terms of the probability density function and the mean value of consensus reliability.
Abstract: The centralized system becomes less efficient, secure, and resilient as the network size and heterogeneity increase due to its inherent single point of failure issues. Distributed consensus mechanisms characterized by decentralization, autonomy, parallelism, and fault-tolerance can meet the increasing demands of safety and security in critical interconnected systems. This article establishes a Node and Link probabilistic failure model in the presence of node and communication link failures for a representative crash fault-tolerant distributed consensus protocol: RAFT. The analytical results in terms of the probability density function and the mean value of consensus reliability are derived. Two important reliability performance indicators, Reliability Gain and Tolerance Gain are proposed to indicate the linear relationship between the consensus reliability and two basic parameters, i.e., the joint failure rate and the maximum number of tolerant faulty nodes, which provide the theoretical guidance for quickly deploying an RAFT system. The special case of a distributed consensus network with already a certain number of failures and its adverse impact are evaluated. The Markov probabilistic models, definitions of Reliability Gain and Tolerance Gain, and the analysis methods proposed in this article can be extended to other consensus mechanisms.
1 citations
TL;DR: In this paper , the problem of jointly optimizing caching and routing decisions with link capacity constraints over an arbitrary network topology is formulated as a continuous diminishing-returns (DR) submodular maximization problem under multiple continuous DR-supermodular constraints, and is NP-hard.
Abstract: We study a cache network in which intermediate nodes equipped with caches can serve requests. We model the problem of jointly optimizing caching and routing decisions with link capacity constraints over an arbitrary network topology. This problem can be formulated as a continuous diminishing-returns (DR) submodular maximization problem under multiple continuous DR-supermodular constraints, and is NP-hard. We propose a poly-time alternating primal-dual heuristic algorithm, in which primal steps produce solutions within $1-\frac{1}{e}$ approximation factor from the optimal. Through extensive experiments, we demonstrate that our proposed algorithm significantly outperforms competitors.
TL;DR: In this article , the authors reviewed the related studies and addressed the problems in the traditional education system with possible solutions, the transition towards smart education, and research challenges in the transition to smart education (i.e., computational and social resistance).
Abstract: IoT is a fundamental enabling technology for creating smart spaces, which can assist the effective face-to-face and online education systems. The transition to smart education (integrating IoT and AI into the education system) is appealing, which has a concrete impact on learners’ engagement, motivation, attendance, and deep learning. Traditional education faces many challenges, including administration, pedagogy, assessment, and classroom supervision. Recent developments in ICT (e.g., IoT, AI and 5G, etc.) have yielded lots of smart solutions for various aspects of life; however, smart solutions are not well integrated into the education system. In particular, the COVID-19 pandemic situation had further emphasized the adoption of new smart solutions in education. This study reviews the related studies and addresses the (i) problems in the traditional education system with possible solutions, (ii) the transition towards smart education, and (iii) research challenges in the transition to smart education (i.e, computational and social resistance). Considering these studies, smart solutions (e.g., smart pedagogy, smart assessment, smart classroom, smart administration, etc.) are introduced to the problems of the traditional system. This exploratory study opens new trends for scholars and the market to integrate ICT, IoT, and AI into smart education.
TL;DR: In this article , the authors reviewed the related studies and addressed the problems in the traditional education system with possible solutions, the transition towards smart education, and research challenges in the transition to smart education (i.e., computational and social resistance).
Abstract: IoT is a fundamental enabling technology for creating smart spaces, which can assist the effective face-to-face and online education systems. The transition to smart education (integrating IoT and AI into the education system) is appealing, which has a concrete impact on learners' engagement, motivation, attendance, and deep learning. Traditional education faces many challenges, including administration, pedagogy, assessment, and classroom supervision. Recent developments in ICT (e.g., IoT, AI and 5G, etc.) have yielded lots of smart solutions for various aspects of life; however, smart solutions are not well integrated into the education system. In particular, the COVID-19 pandemic situation had further emphasized the adoption of new smart solutions in education. This study reviews the related studies and addresses the (i) problems in the traditional education system with possible solutions, (ii) the transition towards smart education, and (iii) research challenges in the transition to smart education (i.e, computational and social resistance). Considering these studies, smart solutions (e.g., smart pedagogy, smart assessment, smart classroom, smart administration, etc.) are introduced to the problems of the traditional system. This exploratory study opens new trends for scholars and the market to integrate ICT, IoT, and AI into smart education.
TL;DR: In this paper , a pebble-game-based memory-efficient optimisation method is proposed for parallel training in IoT networks and edge in B5G networks, which uses operators as scheduling units during training task assignment.
Abstract: Nowadays we are witnessing rapid development of the Internet of Things (IoT), machine learning, and cellular network technologies. They are key components to promote wireless networks beyond 5G (B5G). The plenty of data generated from numerous IoT devices, such as smart sensors and mobile devices, can be utilised to train intelligent models. But it still remains a challenge to efficiently utilise IoT networks and edge in B5G to conduct model training. In this paper, we propose a parallel training method which uses operators as scheduling units during training task assignment. Besides, we discuss a pebble-game-based memory-efficient optimisation in training. Experiments based on various real world network architectures show the flexibility of our proposed method and good performance compared with state of the art.