DMSF: A Dynamic Model Splitting Framework for Edge-Cloud Collaborative Inference

With the popular use of IoT devices, edge computing has been widely applied in the Internet of things (IoT) and regarded as a promising solution for its wide distribution, decentralization, low latency. At the same time, in response to the massive computing data and intelligent requirements of various applications in the IoT, artificial intelligence (AI) technology has also achieved rapid development. As a result, edge intelligence (EI) for the Internet of Things has attracted widespread attention. Driven by the requirement that making full use of data, machine learning (ML) models trained in EI are usually shared. However, there may be some security and privacy issues due to the openness and heterogeneity of edge intelligence. How to ensure flexible data access and data security as well as the accountability for edge nodes and users in EI model sharing have become important issues. In this article, we propose a Ciphertext Policy Attribute Based Proxy Re-encryption (CP-ABPRE) scheme with accountability to address the security and privacy issues in EI model sharing. In our scheme, a user can delegate the access right to others to make model access more flexible. Furthermore, each entity that may need to be held accountable is embedded a unique ID to achieve traceability. Finally, security analysis and performance evaluation are given to prove that our scheme is CPA secure and does not lose much efficiency with more features.

The advent of mobile devices as powerful devices has enabled the development of the sophisticated application. Despite being equipped, many application does not perform well due to the shortage of different capabilities such as computation, memory storage, network bandwidth and battery capacity. Whereas most of the mobile device resources are idle or less utilized unless the user uses them occasionally. This device can be used to create the platform allowing devices within vicinity to form an on-the-fly cloud and collaboratively execute the task which provides various capabilities present in devices as cloud service. The device can send its task to the cloud that would allocate sub-task to a device with the best capability required for executing sub-task, thus improving resource utilization and potentially reduce execution time. We present a ranking algorithm to analyze various capabilities of peer devices and assign a task to the suitable available device using transformed Hungarian algorithm. We propose a local cloud framework with scheduling scheme for assignment of the sub-task to a device with optimized resource use of every capability present in a cloud and distribute workload for executing the resource intensive application.

Multi-protocol label switching (MPLS) technology is useful for IP Virtual Private Networks (IP-VPNs), guaranteeing bandwidth in IP (Internet Procotol) networks, and carrying out traffic engineering with explicit routing. The advantage of MPLS is its high capability to achieve of reliable networks when used with Fast Rerouting. However, Fast Rerouting requires a lot of network resources. This is because, for the rapid recovery of end-to-end communications after dtectoin of failures, secondary LSPs must already have been reserved as detours in case there are failures on primary node-of-node links. The sharing of bandwith among secondary LSPs is thus significant as a way of reducing the usage of network resources when Fast Rerouting is applied.In this paper, we propose a new routing algorithm in which bandwidth is shared among the secondary LSPs for multiple primary LSPs. This algorithm produces efficient network-level LSP designs. Three approaches to the dynamical changing of Open Shortest Path First (OSPF) link-cost metrics are applied in the algorithm. Each approach improves efficiency in the sharing of LSPs. The approaches are (1) the broader distribution of primary LSPs to reduce the need for detours in cases of single failures, (2) the concentration of secondary LSPs on links to increase the possibilities for bandwidth sharing, and (3) the distribution of secondary LSPs that cater to a certain failure, thus increasing the numbers of detouring LSPs which are independent of each other on the respective links. The scheme provides a slight improvement over the results of the conventional Dijkstra-algorithm calculation which is used in conventional OSPF. The proposed algorithms are applied with various network models that have been proposed in IETF Internet drafts, e.g., Haskin, to egress, and Next-Next-Hop; the results are evaluated on the basis of parameters that indicate the consumption of network resources, including the numbers of labels per node, usage of the network's bandwidth capacity, and round-trip times for the secondary LSPs. Application of algorithm (1) led to a 10% lower total LSP cost of the network than the result for the conventional form of bandwidth sharing, i.e., for sharing among the secondary LSPs of a single primary LSP.

Communities residing in proximity to volcanoes face inherent risks from lahars. The eruption of Indonesia's Merapi volcano in 2010 had far-reaching consequences, impacting multiple districts and leading to 386 fatalities, as well as substantial financial losses amounting to approximately US$403 million. In order to bolster disaster response and mitigation efforts, resilience plays a crucial role by considering hazards, risks, vulnerability, and capacity. This research aimed to examine the relationship between disaster health literacy, risk perception, and preparedness, with the ultimate goal of enhancing resilience within a community susceptible to volcanic activities. The study utilized a cross-sectional design, involving 258 participants residing in hazard zone III areas near Yogyakarta's Merapi mountain, selected through proportional clustered sampling. Between January and February 2023, the researchers employed several assessment tools, including the Indonesian versions of the communities advancing resilience toolkit (CART-AS), disaster health literacy mitigation and preparedness questionnaire (DHLQ), disaster risk perception questionnaire (DRPQ), and disaster preparedness questionnaire (DPQ). The collected data underwent analysis utilizing descriptive statistics and the Pearson correlation test. The results unveiled disaster health literacy (β=0.22, p<0.001), risk perception (β=0.26, p<0.001), and preparedness (β=0.38, p<0.001) were all significant predictors of disaster resilience in a volcano-prone community. In conclusion, this study underscores the significant interplay between disaster health literacy, risk perception, and preparedness in relation to disaster resilience. It emphasizes the significance of fortifying disaster health literacy, fostering accurate risk perception, and promoting preparedness measures to enhance overall resilience in the face of disasters. These findings highlight the necessity for targeted interventions and educational programs aimed at augmenting disaster health literacy, improving risk perception, and enhancing preparedness levels within communities, thereby bolstering their resilience and response capabilities effectively.

In-memory key-value stores are often used to speed up Big Data workloads on modern HPC clusters. To maintain their high availability, erasure coding has been recently adopted as a low-cost redundancy scheme instead of replication. Existing erasure-coded update schemes, however, have either low performance or high memory overhead. In this paper, we propose a novel parity logging-based architecture, HybridPL, which creates a hybrid of in-place update (for data and XOR parity chunks) and log-based update (for the remaining parity chunks), so as to balance the update performance and memory cost, while maintaining efficient single-failure repairs. We realize HybridPL as an in-memory key-value store called LogECMem, and further design efficient repair schemes for multiple failures. We prototype LogECMem and conduct experiments on different workloads. We show that LogECMem achieves better update performance over existing erasure-coded update schemes with low memory overhead, while maintaining high basic I/O and repair performance.

With the development of the network, there are more and more multi-thread programs. Multi-thread applications have complex operating states and suffer from data race problem. However, the existing data race detection algorithms are mostly based on dynamic methods, which cause high memory overheads. In this paper, we present DR-Frame, a static data race detector that costs lower memory and time. DR-Frame first compiles the program into an intermediate code file. Then DR-Frame simulates the code file to obtain the read and write information of the program. Finally, DR-Frame checks the data observed and detects data race of the original program. We have validated DR-Frame using three real-world network applications and the PARSEC benchmark suite. The experimental results show that DR-Frame can detect new race cases with low memory overheads, and the time cost is less than executing the program.

Artificial intelligence (AI) is a key enabler of the sixth-generation (6G) mobile communication systems. Traditional AI infrastructures rely on cloud services, leading to high costs and data security concerns. To address the limitations of AI centralization, this thesis explores blockchain, a distributed ledger technology, to decentralize AI to the 6G network edge. To improve the quality of service (QoS) in edge intelligence systems, we optimize the training and inference processes, considering unique challenges such as disparate data quality, resource limitations, and data distribution shifts. First, we investigate a collaborative decentralized machine learning (DML) approach. For the first time, blockchain is used as a time-dependent incentive mechanism with a diversified reward model. Instead of fine-tuning AI models, we focus on assessing and conditioning data at the 6G edge. A decentralized data valuation protocol is introduced atop blockchain to incentivize contributors and disincentivize free-riders. Simulation results show that optimizing information loss and the marginal utility of individual data samples is key to improving QoS in collaborative edge intelligence systems. Second, we focus on blockchain-enabled DML with enhanced security and reduced latency. We propose a novel proof-of-useful-work protocol to utilize edge computing resources efficiently. Moreover, we apply mean-field theory and introduce a novel aggregation protocol named Corrected Krum to mitigate the negative impact of data poisoning attacks. To minimize per-round latency, the optimization problem is formulated as a multi-way number partitioning challenge. Extensive simulations demonstrate that our blockchain approach enhances reliability and QoS for edge AI systems in diverse data environments. Finally, we optimize edge AI deployment. To address data distribution shifts during model inference, a blockchain-based model selection approach is developed for test-time adaptation. Blockchain consensus is applied for the first time to deliver cost-effective inference for resource-constrained edge devices. Our blockchain approaches enable DML to adapt to data distribution shifts with low latency and superior prediction accuracy. Furthermore, we decentralize multi-modal large AI model inference to the 6G edge. An online bipartite matching protocol is proposed for rapid resource allocation during peak service hours. Experimental results confirm blockchain’s potential for reliable and collaborative edge intelligence with improved QoS.

Recently, the concept of fog computing which aims at providing time-sensitive data services has become popular. In this model, computation is performed at the edge of the network instead of sending vast amounts of data to the cloud. Thus, fog computing provides low latency, location awareness to end users, and improves quality-of-services (QoS). One key feature in this model is the designing of payment plan from network operator (NO) to fog nodes (FN) for the rental of their computing resources, such as computation capacity, spectrum, and transmission power. In this paper, we investigate the problem of how to design the efficient payment plan to maximize the NO's revenue while maintaining FN's incentive to cooperate through the moral hazard model in contract theory. We propose a multi-dimensional contract which considers the FNs' characteristics such as location, computation capacity, storage, transmission bandwidth, and etc. First, a contract which pays the FNs by evaluating the resources they have provided from multiple aspects is proposed. Then, the utility maximization problem of the NO is formulated. Furthermore, we use the numerical results to analyze the optimal payment plan, and compare the NO's utility under different payment plans.

SummaryThe problem of flash data dissemination refers to transmitting time‐critical data to a large group of distributed receivers in a timely manner, which widely exists in many mission‐critical applications and Web services. However, existing approaches for flash data dissemination fail to ensure the timely and efficient transmission, because of the unpredictability of the dissemination process. Overlay routing has been widely used as an efficient routing primitive for providing better end‐to‐end routing quality by detouring inefficient routing paths in the real networks. To improve the predictability of the flash data dissemination process, we propose a bandwidth and latency sensitive overlay routing approach named BLOR, by optimizing the overlay routing and avoiding inefficient paths in flash data dissemination. BLOR tries to select optimal routing paths in terms of network latency, bandwidth capacity, and available bandwidth in nature, which has never been studied before. Additionally, a location‐aware unstructured overlay topology construction algorithm, an unbiased top‐kdominance model, and an efficient semi‐distributed information management strategy are proposed to assist the routing optimization of BLOR. Extensive experiments have been conducted to verify the effectiveness and efficiency of the proposals with real‐world data sets. Copyright © 2014 John Wiley & Sons, Ltd.

There is an increasing interest in executing complex analyses over large graphs, many of which require processing a large number of multi-hop neighborhoods or subgraphs. Examples include ego network analysis, motif counting, finding social circles, personalized recommendations, link prediction, anomaly detection, analyzing influence cascades, and others. These tasks are not well served by existing vertex-centric graph processing frameworks, where user programs are only able to directly access the state of a single vertex at a time, resulting in high communication, scheduling, and memory overheads in executing such tasks. Further, most existing graph processing frameworks ignore the challenges in extracting the relevant portions of the graph that an analysis task is interested in, and loading those onto distributed memory. This paper introduces NScale, a novel end-to-end graph processing framework that enables the distributed execution of complex subgraph-centric analytics over large-scale graphs in the cloud. NScale enables users to write programs at the level of subgraphs rather than at the level of vertices. Unlike most previous graph processing frameworks, which apply the user program to the entire graph, NScale allows users to declaratively specify subgraphs of interest. Our framework includes a novel graph extraction and packing (GEP) module that utilizes a cost-based optimizer to partition and pack the subgraphs of interest into memory on as few machines as possible. The distributed execution engine then takes over and runs the user program in parallel on those subgraphs, restricting the scope of the execution appropriately, and utilizes novel techniques to minimize memory consumption by exploiting overlaps among the subgraphs. We present a comprehensive empirical evaluation comparing against three state-of-the-art systems, namely Giraph, GraphLab, and GraphX, on several real-world datasets and a variety of analysis tasks. Our experimental results show orders-of-magnitude improvements in performance and drastic reductions in the cost of analytics compared to vertex-centric approaches.

Partitioning links rather than nodes is effective in overlapping community detection (OCD) on complex networks. However, it consumes high CPU and memory overheads because the volume of links is huge especially when the network is rather complex. In this paper, we proposes a symmetric non-negative matrix factorization (SNMF) based link partition method called SNMF-Link to overcome this deficiency. In particular, SNMFLink represents data in a lower-dimensional space spanned by the node-link incidence matrix. By solving a lighter SNMF problem, SNMF-Link learns the clustering indicators of each links. Since traditional multiplicative update rule (MUR) based optimization algorithm for SNMF suffers from slow convergence, we applied the augmented Lagrangian method (ALM) to efficiently optimize SNMF. Experimental results show that SNMF-Link is much more efficient than the representative clustering algorithms without reducing the OCD performance.

Checkpoint replication is a prevalent way of maintaining virtual machine availability in the presence of host failures. Since checkpoint replication can impose heavy load on network resources, checkpoint compression has been suggested to reduce network usage. This paper presents the first detailed evaluation and characterization of the effectiveness and overheads of checkpoint compression methods for various workloads frequently seen in high-availability systems. We propose a lightweight compression method that exploits similarities in checkpoints to eliminate redundant network traffic, and compare it with two well-known methods, gzip and delta compression. Our results show that gzip and delta compression reduce network traffic significantly for various workloads, but incur high CPU and memory overheads, respectively. The proposed similarity compression is most effective for VM clusters running homogeneous workloads, while using both CPU and memory efficiently. Based on our extensive evaluation, we suggest guidelines for selecting and using these compression methods.

AugFCOS: Augmented fully convolutional one-stage object detection network

The rapid growth of video transmission and the increasing demand for high-definition, multi-display, and the wide-area video services in recent years. It resulted in the need for robust technologies to cover these requirements. RLNC is one of the most promising methods for video delivery, improving throughput, utilizing bandwidth capacity, and saving High reliability and low latency. In this paper, we discuss the possibility of applying the RLNC algorithm to video transmission and the benefit of the features of its previously mentioned. We design a proposed model using Matlab consisting of three stages: encoder in the source node, recorder in the intermediate node, and decoder in the destination node. From this model, we got a zero-bit error rate BER between the frames to transmit and receive end, throughput is increased by 25% from previous, and increasing bandwidth capacity utilization by sending more than one packet in the transmission process, thus reducing the number of transmission times.

At the confluence of two great paradigms such as Edge Computing and Artificial Intelligence, Edge Intelligence arises. This new concept is about the smart exploitation of Edge Computing by bringing together reasoning and learning by Artificial Intelligence algorithms and the sensors/actuators computing capabilities. Security is the third paradigm that must join the team in order to have resilient and reliable systems to be used in real-world applications and use cases. Hence, smartness is, in this context, a puzzle of several independent pieces which, once fitted, can derived unprecedented benefits: a) security, b) low communication latency and network load, c) cost and energy saving and d) scalability by means of resource virtualization close to the IoT data generators (IoT devices). In fact, by paying exclusive attention to some of those main pillars and, therefore, disregarding others, edge computation once in operation often suffers from bad performance, unforeseen events or does not exploit the enormous potential that should be unlocked if a proper and complete specification had been laid down. With all this in mind, this work provides a technical review of the available and up-to-date frameworks to implement secure Edge Intelligence, pinpoints the most relevant unfilled gaps (strengths and weaknesses) and, last but nos least, includes challenges and future research lines as a result of our exploration.