The proliferation of the Internet of Things (IoT) has increasingly demanded intimacy between cloud services and end-users. This has incentivised extending cloud resources to the edge in what is deemed fog computing. The latter is manifesting as an ecosystem of connected clouds, geo-dispersed and of diverse capacities. In such conditions, workload allocation to fog services becomes a non-trivial challenge due to the complexity of trade-offs. Users' demand at the edge is highly diverse, which does not lend itself to straightforward resource planning. Conversely, running services at the edge may leverage proximity, but it comes at higher operational cost let alone rapidly increasing the risk of straining sparse resources. Consequently, there is a need for intelligent yet scalable allocation solutions that counter the adversity of demand at the edge, while efficiently distributing load between the edge and farther clouds. Machine learning is increasingly adopted in resource planning. However, besides privacy concerns, central learning is highly demanding, both computationally and in data supply. Instead, this paper proposes a federated deep reinforcement learning system, based on deep Q-learning network (DQN), for workload distribution in a fog ecosystem. The proposed solution adapts a DQN to optimize local workload allocations, made by single gateways. Federated learning is incorporated to allow multiple gateways in a network to collaboratively build knowledge of users' demand. This is leveraged to establish consensus on the fraction of workload allocated to different fog nodes, using lower data supply and computation resources. The system performance is evaluated using realistic demand set from Google Cluster Workload Traces 2019. Evaluation results show over 50% reduction in failed allocations when distributing users over larger number of gateways, given fixed number of fog nodes. The results further illustrate the trade-offs between performance and cost under different conditions.

With increasing amounts of data coming fromconnecting progressively more devices, new machine learningmodels have risen. For wireless networks the idea of using adistributed approach to machine learning has gained increasingpopularity, where all nodes in the network participate in creatinga global machine learning model by training with the localdata stored at each node, an example of this approach is calledfederated learning. However, traditional communication protocolshave been proven inefficient. This opens up opportunities todesign new machine-learning specific communication schemes.The concept ofOver-the-air computationis built on the fact thata wireless communication channel can naturally compute somelinear functions, for instance the sum. If all nodes in a networktransmits simultaneously to a server, the signals are aggregatedbefore reaching the server. / I takt med denökande datamängden frånallt fler uppkopplade enheter har nya modeller för mask-ininlärning dykt upp. För trådlösa nätverk har idén att appliceradecentraliserade maskininlärnings modellerökat i popularitet,där alla noder i nätverket bidrar till en global maskininlärningsmodell genom att träna på den data som finns lokalt på varjenod. Ett exempel på en sådan metodärFederated Learning.Traditionella metoder för kommunikation har visat sig varaineffektiva vilket öppnar upp möjligheten för att designa nyamaskininlärningsspecifika kommunikationsscheman. Konceptetover-the-air computationutnyttjar det faktum att en trådlöskommunikationskanal naturligt kan beräkna vissa funktioner,som exempelvis en summa. Om alla noder i nätverket sändertill en server samtidigt aggregeras signalerna genom interferensinnan de når servern. / Kandidatexjobb i elektroteknik 2020, KTH, Stockholm

The Internet of Vehicles (IoV) holds immense potential for revolutionizing transporta- tion systems by facilitating seamless vehicle-to-vehicle and vehicle-to-infrastructure communication. However, challenges such as congestion, pollution, and security per- sist, particularly in rural areas with limited infrastructure. Existing centralized solu- tions are impractical in such environments due to latency and privacy concerns. To address these challenges, we propose a decentralized clustering algorithm enhanced with Federated Deep Reinforcement Learning (FDRL). Our approach enables low- latency communication, competitive packet delivery ratios, and cluster stability while preserving data privacy. Additionally, we introduce a trust-based security framework for IoV environments, integrating a central authority and trust engine to establish se- cure communication and interaction among vehicles and infrastructure components. Through these innovations, we contribute to safer, more efficient, and trustworthy IoV deployments, paving the way for widespread adoption and realizing the transfor- mative potential of IoV technologies.

This thesis presents an innovative strategy for distributed Graph Neural Network (GNN) training, leveraging a peer-to-peer network of heterogeneous edge devices interconnected through a Distributed Hash Table (DHT). As GNNs become increasingly vital in analyzing graph-structured data across various domains, they pose unique challenges in computational demands and privacy preservation, particularly when deployed for training on edge devices like smartphones. To address these challenges, our study introduces the Adaptive Load- Balanced Partitioning (ALBP) technique in the GraphDHT system. This approach optimizes the division of graph datasets among edge devices, tailoring partitions to the computational capabilities of each device. By doing so, ALBP ensures efficient resource utilization across the network, significantly improving upon traditional participant selection strategies that often overlook the potential of lower-performance devices. Our methodology's core is weighted graph partitioning and model aggregation in GNNs, based on partition ratios, improving training efficiency and resource use. ALBP promotes inclusive device participation in training, overcoming computational limits and privacy concerns in large-scale graph data processing. Utilizing a DHT-based system enhances privacy in the peer-to-peer setup. The GraphDHT system, tested across various datasets and GNN architectures, shows ALBP's effectiveness in distributed GNN training and its broad applicability in different domains and structures. This contributes to applied machine learning, especially in optimizing distributed learning on edge devices. / Master of Science / Graph Neural Networks (GNNs) are a type of machine learning model that focuses on analyzing data structured like a network, such as social media connections or biological systems. These models can help identify patterns and make predictions in various tasks, but training them on large-scale datasets can require significant computing power and careful handling of sensitive data. This research proposes a new method for training GNNs on small devices, like smartphones, by dividing the data into smaller pieces and using a peer-to-peer (p2p) network for communication between devices. This approach allows the devices to work together and learn from the data while keeping sensitive information private. The main contributions of this research are threefold: (1) examining existing ways to divide network data and how they can be used for training GNNs on small devices, (2) improving the training process by creating a localized, decentralized network of devices that can communicate and learn together, and (3) testing the method on different types of datasets and GNN models, showing that it works well across a variety of situations. To sum up, this research offers a novel way to train GNNs on small devices, allowing for more efficient learning and better protection of sensitive information.

In recent years, Internet-of-Things (IoT) devices generate a large amount of personal data. However, due to the privacy concern, collecting the private data in cloud centers for training Machine Learning (ML) models becomes unrealistic. To address this problem, Federated Learning (FL) is proposed. Yet, central bottleneck has become a severe concern since the central node in traditional FL is responsible for the communication and aggregation of mil- lions of edge devices. In this paper, we propose Scalable Federated Learning via Distributed Hash Table Based Overlays for network (Scaled) to conduct multiple concurrently running FL-based applications over edge networks. Specifically, Scaled adopts a fully decentral- ized multiple-master and multiple-slave architecture by exploiting Distributed Hash Table (DHT) based overlay networks. Moreover, Scaled improves the scalability and adaptability by involving all edge nodes in training, aggregating, and forwarding. Overall, we make the following contributions in the paper. First, we investigate the existing FL frameworks and discuss their drawbacks. Second, we improve the existing FL frameworks from centralized master-slave architecture by using DHT-based Peer-to-Peer (P2P) overlay networks. Third, we implement the subscription-based application-level hierarchical forest for FL training. Finally, we demonstrate Scaled's scalability and adaptability over large scale experiments. / Master of Science / In recent years, Internet-of-Things (IoT) devices generate a large amount of personal data. However, due to privacy concerns, collecting the private data in central servers for training Machine Learning (ML) models becomes unrealistic. To address this problem, Federated Learning (FL) is proposed. In traditional ML, data from edge devices (i.e. phones) should be collected to the central server to start model training. In FL, training results, instead of the data, are collected to perform training. The benefit of FL is that private data can never be leaked during the training. However, there is a major problem in traditional FL: a single point of failure. When power to a central server goes down or the central server is disconnected from the system, it will lose all the data. To address this problem, Scaled: Scalable Federated Learning via Distributed Hash Table Based Overlays is proposed. Instead of having one powerful main server, Scaled launches many different servers to distribute the workload. Moreover, since Scaled is able to build and manage multiple trees at the same time, it allows multi-model training.

The future quantum Internet (QI) will transform today's communication networks and user experiences by providing unparalleled security levels, superior quantum computational powers, along with enhanced sensing accuracy and data processing capabilities. These features will be enabled through applications like quantum key distribution (QKD) and quantum machine learning (QML). Towards enabling these applications, the QI requires the development of global quantum communication networks (QCNs) that enable the distribution of entangled resources between distant nodes. This dissertation addresses two major challenges facing QCNs, which are the scalability and coverage of their architectures, and the efficiency of their operations. Additionally, the dissertation studies the near-term deployment of QML applications over today's noisy quantum devices, essential for realizing the future QI. In doing so, the scalability and efficiency challenges facing the different QCN elements are explored, and practical noise-aware and physics-informed approaches are developed to optimize the QCN performance given heterogeneous quantum application-specific quality of service (QoS) user requirements on entanglement rate and fidelity. Towards achieving this goal, this dissertation makes a number of key contributions. First, the scaling limits of quantum repeaters is investigated, and a holistic optimization framework is proposed to optimize the geographical coverage of quantum repeater networks (QRNs), including the number of quantum repeaters, their placement and separating distances, quantum memory management, and quantum operations scheduling. Then, a novel framework is proposed to address the scalability challenge of free-space optical (FSO) quantum channels in the presence of blockages and environmental effects. Particularly, the utilization of a reconfigurable intelligent surface (RIS) in QCNs is proposed to maintain a line-of-sight (LoS) connection between quantum nodes separated by blockages, and a novel analytical model of quantum noise and end-to-end (e2e) fidelity in such QCNs is developed. The results show enhanced entangled state fidelity and entanglement distribution rates, improving user fairness by around 40% compared to benchmark approaches. The dissertation then investigates the efficiency challenges in a practical use-case of QCNs with a single quantum switch (QS). Particularly, the average quantum memory noise effects are analytically analyzed and their impacts on the allocation of entanglement generation sources and minimization of entanglement distribution delay while optimizing QS entanglement distillation operations are investigated. The results show an enhanced e2e fidelity and a minimized e2e entanglement distribution delay compared to existing approaches, and a unique capability of satisfying all users QoS requirements. This QCN architecture is scaled up with multiple QSs serving heterogeneous user requests, necessary for scalable quantum applications over the QI. Here, a novel efficient matching theory-based framework for optimizing the request-QS association in such QCNs while managing quantum memories and optimizing QS operations is proposed. Finally, after scaling QCNs and ensuring their efficient operations, the dissertation proposes novel distributed QML frameworks that can leverage both classical networks and QCNs to enable collaborative learning between today's noisy quantum devices. In particular, the first quantum federated learning (QFL) frameworks incorporating different quantum neural networks and leveraging quantum and classical data are developed, and the first publicly available federated quantum dataset is introduced. The results show enhanced performance and reductions in the communication overhead and number of training epochs needed until convergence, compared to classical counterpart frameworks. Overall, this dissertation develops robust frameworks and algorithms that advance the theoretical understanding of QCNs and offers practical insights for the future development of the QI and its applications. The dissertation concludes by analyzing some open challenges facing QCNs and proposing a vision for physics-informed QCNs, along with important future directions. / Doctor of Philosophy / In today's digital age, we are generating vast amounts of data through videos, live streams, and various online activities. This explosion of data brings not only incredible opportunities for innovation but also heightened security concerns. The current Internet infrastructure struggles to keep up with the demand for speed and security. In this regard, the quantum Internet (QI) emerges as a revolutionary technology poised to make the communication and data sharing processes faster and more secure than ever before. The QI requires the development of quantum communication networks (QCNs) that will be seamlessly integrated with today's existing communication systems that form today's Internet. This way, the QI enables ultra-secure communication and advanced computing applications that can transform various sectors, from finance to healthcare. However, building such global QCNs, requires overcoming significant challenges, including the sensitive nature and limitations of quantum devices. In this regard, the goal of this dissertation is to develop scalable and efficient QCNs that overcome the different challenges facing different QCN elements and enable a wide coverage and robust performance towards realizing the QI at a global scale. Simultaneously, machine learning (ML), which is driving significant advancements and transforming industries in today's world. Here, quantum technologies are anticipated to make a breakthrough in ML through quantum machine learning (QML) models that can handle today's large and complex data. However, quantum computers are still limited in scale and efficiency, often being noisy and unreliable. Throughout this dissertation, these limitations of QML are addressed by developing frameworks that allow multiple quantum computers to work together collaboratively in a distributed manner over classical networks and QCNs. By leveraging distributed QML, it is possible to achieve remarkable advancements in privacy and data utilization. For instance, distributed QML can enhance navigation systems by providing more accurate and secure route planning or revolutionize healthcare by enabling secure and efficient analysis of medical data. In summary, this dissertation addresses the critical challenges of building scalable and efficient QCNs to support the QI and develops distributed QML frameworks to enable near-term utilization of QML in transformative applications. By doing so, it paves the way for a future where quantum technology is integral to our daily lives, enhancing security, efficiency, and innovation across various domains.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第25427号 / 情博第865号 / 京都大学大学院情報学研究科社会情報学専攻 / (主査)教授 伊藤 孝行, 教授 黒田 知宏, 教授 岡部 寿男, 吉川 正俊(京都大学 名誉教授) / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

<p dir="ltr">Sparse basis recovery is an important learning problem when the number of model dimensions (<i>p</i>) is much larger than the number of samples (<i>n</i>). However, there has been little work that studies sparse basis recovery in the Federated Learning (FL) setting, where the Differential Privacy (DP) of the client data must also be simultaneously protected. Notably, the performance guarantees of existing DP-FL algorithms (such as DP-SGD) will degrade significantly when the system is ill-determined (i.e., <i>p >> n</i>), and thus they will fail to accurately learn the true underlying sparse model. The goal of my thesis is therefore to develop DP-FL sparse basis recovery algorithms that can recover the true underlying sparse basis provably accurately even when <i>p >> n</i>, yet still guaranteeing the differential privacy of the client data.</p><p dir="ltr">During my PhD studies, we developed three DP-FL sparse basis recovery algorithms for this purpose. Our first algorithm, SPriFed-OMP, based on the Orthogonal Matching Pursuit (OMP) algorithm, can achieve high accuracy even when <i>n = O(\sqrt{p})</i> under the stronger Restricted Isometry Property (RIP) assumption for least-square problems. Our second algorithm, Humming-Bird, based on a carefully modified variant of the Forward-Backward Algorithm (FoBA), can achieve differentially private sparse recovery for the same setup while requiring the much weaker Restricted Strong Convexity (RSC) condition. We further extend Humming-Bird to support loss functions beyond least-square satisfying the RSC condition. To the best of our knowledge, these are the first DP-FL results guaranteeing sparse basis recovery in the <i>p >> n</i> setting.</p>

The distributed and privacy sensitive nature of cellular networks make them strong candidates for optimization using Federated Learning, but this exposes them to a problem inherent to the learning paradigm: performance inequality due to heterogeneous client data distributions. The prevailing approach of enforcing uniform client performance ignores client-specific performance limitations due to different levels of irreducible uncertainty present in their data, resulting in deteriorated network performance. To address this issue, this thesis introduces two novel federated algorithms designed to enhance learning efficiency and ensure fairness in the presence of heteroscedastic noise, reflecting the distributive justice principles of utilitarianism and equality. Under these circumstances, the proposed algorithms are shown to significantly improve overall performance and performance fairness. The deployment of these algorithms promises a dual benefit: enhancement in network performance and a fairer distribution of service quality for end users.

Decentralized Machine Learning could address some problematic facets with Federated Learning. There is no central server acting as an arbiter of whom or what may benefit from Machine Learning models created by the vast amount of data becoming available in recent years. It could also increase the reliability and scalability of Machine Learning systems thereby drawing the benefit of having more data accessible. Gossip Learning is such a protocol, but has primarily been designed with linear models in mind. How does Gossip Learning perform when training Deep Neural Networks? Could it be a viable alternative to Federated Learning? In this thesis, we implement Gossip Learning using two different model merging strategies. We also design and implement two extensions to this protocol with the goal of achieving higher performance when training under churn. The training methods are compared on two tasks: image classification on the Federated Extended MNIST dataset and time- series forecasting on the NN5 dataset. Additionally, we also run an experiment where learners churn, alternating between being available and unavailable. We find that Gossip Learning performs slightly better in settings where learners do not churn but is vastly outperformed in the setting where they do. / Decentraliserad Maskinginlärning kan lösa några problematiska aspekter med Federated Learning. Det finns ingen central server som agerar som domare för vilka som får gagna av Maskininlärningsmodellerna skapad av den stora mäng data som blivit tillgänglig på senare år. Det skulle också kunna öka pålitligheten och skalbarheten av Maskininlärningssystem och därav dra nytta av att mer data är tillgänglig. Gossip Learning är ett sånt protokoll, men det är primärt designat med linjära modeller i åtanke. Hur presterar Gossip Learning när man tränar Djupa Neurala Nätverk? Kan det vara ett möjligt alternativ till Federated Learning? I det här exjobbet implementerar vi Gossip Learning med två olika modelsammanslagningstekniker. Vi designar och implementerar även två tillägg till protokollet med målet att uppnå bättre prestanda när man tränar i system där noder går ner och kommer up. Träningsmetoderna jämförs på två uppgifter: bildklassificering på Federated Extended MNIST datauppsättningen och tidsserieprognostisering på NN5 datauppsättningen. Dessutom har vi även experiment då noder alternerar mellan att vara tillgängliga och otillgängliga. Vi finner att Gossip Learning presterar marginellt bättre i miljöer då noder alltid är tillgängliga men är kraftigt överträffade i miljöer då noder alternerar mellan att vara tillgängliga och otillgängliga.