Collaborative data sharing underlies applications in systems such as online social networks and cloud storage. A central provider hosts shared data, e.g., a Facebook group page, and provides sharing users with read/wr...
详细信息
Collaborative data sharing underlies applications in systems such as online social networks and cloud storage. A central provider hosts shared data, e.g., a Facebook group page, and provides sharing users with read/write access according to user-defined settings. Historical incidents prove that data storage centralization enables a corrupted provider to censor or diverge users' views of the shared data. Such misbehavior is hardly detectable as users update the shared data unbeknown to each other. Existing solutions suggest server-side data replication, which is storage inefficient, or users' out-of-band communication, which is communication intensive. Addressing these inefficiencies, we propose Integrita, a data-sharing mechanism that protects users' view-consistency needless to out-of-band communication. We present a novel distributed history tree algorithm to partition the shared data among N storage providers, N - 1 of which are Byzantine faulty and colluding. Our data partitioning solution reduces storage overhead by a multiplicative factor of N and allows provable detection of server-side equivocation and identification of corrupted servers. We introduce and achieve anew consistency level, named q-detectable consistency, where users' views inconsistency cannot remain undetected for more than q updates, q being the system parameters' function. Without loss of generality, we use online social networks as a case study to deploy Integrita and supply performance/numerical analysis accordingly.
The widespread use of the Internet and digital services has significantly increased data collection and processing. Critical domains like healthcare rely on this data, but privacy and security concerns limit its usabi...
详细信息
The widespread use of the Internet and digital services has significantly increased data collection and processing. Critical domains like healthcare rely on this data, but privacy and security concerns limit its usability, constraining the performance of intelligent systems, particularly those leveraging Neural networks (NNs). NNs require high-quality data for optimal performance, but existing privacy-preserving methods, such as Federated Learning and Differential Privacy, often degrade model accuracy. While Homomorphic Encryption (HE) has emerged as a promising alternative, existing HE-based methods face challenges in computational efficiency and scalability, limiting their real-world application. To address these issues, we introduce ENNigma, a novel framework employing state-of-the-art Fully Homomorphic Encryption techniques. This framework introduces optimizations that significantly improve the speed and accuracy of encrypted NN operations. Experiments conducted using the CIC-DDoS2019 dataset - a benchmark for distributed Denial of Service attack detection - demonstrate ENNigma's effectiveness. A classification performance with a maximum relative error of 1.01% was achieved compared to non-private models, while reducing multiplication time by up to 59% compared to existing FHE-based approaches. These results highlight ENNigma's potential for practical, privacy-preserving neural network applications.
As model sizes in machine learning continue to scale, distributed training is necessary to accommodate model weights within each device and to reduce training time. However, this comes with the expense of increased co...
详细信息
Recent years have witnessed significant advancements in Artificial Intelligence (AI), particularly with the rise of Deep Neural networks fueled by large datasets and increased model complexity. However, the demand for...
详细信息
ISBN:
(纸本)9798400704130
Recent years have witnessed significant advancements in Artificial Intelligence (AI), particularly with the rise of Deep Neural networks fueled by large datasets and increased model complexity. However, the demand for substantial computational resources poses challenges in centralized data scenarios. Edge Intelligence (EI), combining Edge Computing and AI, emerges as a transformative solution for decentralized learning, crucial in the era of IoT proliferation. While Federated Learning (FL) has been a prominent paradigm in decentralized learning, its limitations have prompted researchers to explore alternative solutions using Knowledge Distillation (KD) as a basis. The purpose of this Ph.D. research is to explore KD as a new paradigm for decentralized learning, contribute to enhancing its performance, and study the trade-off between FL and KD in terms of efficiency and effectiveness to identify best practices and insights in EI environments.
Communication needs in avionics and transportation have radically changed over the recent years. Traditionally, the underlying hard real-time networks were designed in a centralized way, focusing on redundancy and iso...
详细信息
ISBN:
(纸本)9798350371291;9798350371284
Communication needs in avionics and transportation have radically changed over the recent years. Traditionally, the underlying hard real-time networks were designed in a centralized way, focusing on redundancy and isolation. Today, real-time communication is ubiquitous, from large airplanes to small vehicles. The associated networks must support a wide range of applications, and large amounts of data. Centralized approaches from the avionics domain, e.g., AFDX, are too costly, too heavyweight, and not flexible enough for these applications. In this paper we explore a new distributed network architecture designed to support jumbo airliners, but also small aircraft and drones. Communication redundancy is achieved using redundant paths, which have to be adapted and optimized to the application. The main challenge then is to build an optimized network configuration ensuring safety, fault tolerance, timing, and performance of both critical, and non-critical communication. Minimizing volume and weight of the equipment is also mandatory. Since the solution space is too large to be explored in reasonable time, we propose a genetic algorithm. Our experiments show that our algorithm converges quickly and offers solutions of excellent quality. The computed solutions are in the top 2% among the best solutions obtained using an exhaustive exploration. Our approach thus enables system engineers to quickly explore and choose very good solution for their systems.
In this study, we explore the integration of satellites with ground-based communication networks. Specifically, we analyze downlink data transmission from a constellation of satellites to terrestrial users and address...
详细信息
ISBN:
(纸本)9798350362527;9798350362510
In this study, we explore the integration of satellites with ground-based communication networks. Specifically, we analyze downlink data transmission from a constellation of satellites to terrestrial users and address the issue of delayed channel state information (CSI). The satellites cooperate in data transmission within a cluster to create a unified, distributed massive multiple input, multiple output (MIMO) system. The CSI used for this process is inherently outdated, particularly due to the delay from the most distant satellite in the cluster. Therefore, in this paper, we develop a precoding strategy that leverages the long-term characteristics of CSI uncertainty to compensate for the undesirable impact of these unavoidable delays. Our proposed method is computationally efficient and particularly effective in lower frequency bands. As such, it holds significant promise for facilitating the integration of satellite and terrestrial communication, especially within frequency bands of up to 1 GHz.
The proceedings contain 3 papers. The topics discussed include: analysis and evaluation of load management strategies in a decentralized FaaS environment: a simulation-based framework;live migration of multi-container...
ISBN:
(纸本)9798400706479
The proceedings contain 3 papers. The topics discussed include: analysis and evaluation of load management strategies in a decentralized FaaS environment: a simulation-based framework;live migration of multi-container Kubernetes pods in multi-cluster serverless edge systems;and comparing actor-critic and neuroevolution approaches for traffic offloading in FaaS-powered edge systems.
Increasingly complex and diverse deep neural network (DNN) models necessitate distributing the execution across multiple devices for training and inference tasks, and also require carefully planned schedules for perfo...
详细信息
ISBN:
(纸本)9798350393132;9798350393149
Increasingly complex and diverse deep neural network (DNN) models necessitate distributing the execution across multiple devices for training and inference tasks, and also require carefully planned schedules for performance. However, existing practices often rely on predefined schedules that may not fully exploit the benefits of emerging diverse model-aware operator placement strategies. Handcrafting high-efficiency schedules can be challenging due to the large and varying schedule space. This paper presents Tessel, an automated system that searches for efficient schedules for distributed DNN training and inference for diverse operator placement strategies. To reduce search costs, Tessel leverages the insight that the most efficient schedules often exhibit repetitive pattern across different data inputs. This leads to a two-phase approach: repetend construction and schedule completion. By exploring schedules for various operator placement strategies, Tessel significantly improves both training and inference performance. Experiments with representative DNN models demonstrate that Tessel achieves up to 5.5x training performance speedup and up to 38% inference latency reduction.
The combined approach proposes the use of PBFT and Raft to ensure data consistency and fault tolerance in the system, and also integrates recurrent neural networks to analyze and predict the behavior of nodes in the n...
详细信息
ISBN:
(数字)9783031651540
ISBN:
(纸本)9783031651533;9783031651540
The combined approach proposes the use of PBFT and Raft to ensure data consistency and fault tolerance in the system, and also integrates recurrent neural networks to analyze and predict the behavior of nodes in the network. RNNs can be used to detect anomalies, predict system load, and analyze time series data related to node operation. The proposed combined approach opens up new prospects for the development of distributedsystems, increasing their reliability, fault tolerance and adaptability to changing conditions. Further research in this direction could lead to more efficient and secure distributedsystems that can efficiently handle complex real-world scenarios.
Sensor networks (SN) could be defined as networks of autonomous devices that can sense and/or act on physical or environmental conditions cooperatively. In sensor networks, data is typically sensed and sent to an aggr...
详细信息
ISBN:
(纸本)9781665451093
Sensor networks (SN) could be defined as networks of autonomous devices that can sense and/or act on physical or environmental conditions cooperatively. In sensor networks, data is typically sensed and sent to an aggregator that will process it with local AI or send it to a cloud with larger AI. This centralized architecture has drawbacks such as the aggregator having to receive and process a potentially huge amount of data, which results in power consumption that can be significant. In addition, the transmission of all the sensor data results in extra consumption of energy due to communication. To reduce the impact of this last point, on-edge computing allows data to be preprocessed at the sensor level. It is often used in the architecture of distributed sensor networks in which each node receives data from other nodes in the network and processes it with its local data. In this work, a distributed sensor network model aims to solve this problem while reducing the impact of data transmission on energy consumption. The proposed model is able to reduce the number of transmitted bits per node by 90% in a nodeto-node directed communication scenario. It is also capable of working with different AI paradigms depending on the required balance between energy consumption, application configurability, and accuracy. Finally, this model is capable of converging even in a network without complete interconnections between nodes.
暂无评论