Distributed Ledger Technology (DLT), including blockchain, is increasingly used within industry ecosystems to create a high-integrity, single source of truth of shared data and business processes across diverse parties. A major challenge in adopting DLT is the conflicting demands on data transparency for improved integrity, against hiding commercially sensitive data. To address this, some industry ecosystems use multiple ledgers shared only between relevant parties rather than using a single distributed ledger across the entire ecosystem. The design problem for this is: What parties should share which ledgers, and what data should be on those ledgers? In this article, we propose a method employing a Design Structure Matrix (DSM) and Domain Mapping Matrix (DMM) to derive candidate shared ledger combinations for industry ecosystems. The method also indicates, for certain data, centralized web services or point-to-point messages may be more suitable than shared ledgers. We discuss our experiences with applying this method while developing a prototype for an agricultural traceability platform. We also present a genetic-algorithm-based DSM and DMM clustering technique to derive candidate shared ledger combinations.

Artificial intelligence of things (AIoT) is an emerging paradigm integrating artificial intelligence (AI) technologies within the Internet of Things (IoT) paradigm. However, deploying deep-learning models on IoT devices is challenging due to their inherent computational, communications, and security constraints. To address these challenges, we propose Flexible Neuron Selection (FlexNS), a computation- and communication-efficient personalised multi-task transfer learning framework for AIoT.FlexNSenables IoT devices to train their private task-specific shallow models by leveraging a multi-task, deep-learning model pre-trained by a cloud server.FlexNSsignificantly reduces IoT devices’ computational and communications resource demands by selecting a subset of neurons in an early layer of the server’s public model to be connected to the private models of multiple IoT devices. The neurons need to be carefully selected to ensure effective and efficient knowledge transfer to the fine-tuned private models tailored to each IoT device’s specific task. Experimental results show thatFlexNS-based private models achieve 104.3% and 98.4% model accuracy compared to the public model for two datasets on network intrusion detection and image classification tasks, with 99.5% and 98.0% reduction in training and inference time.

The vulnerability of neural networks to adversarial perturbations has necessitated formal verification techniques that can rigorously certify the quality of neural networks. As the state-of-the-art, branch-and-bound (BaB) is a "divide-and-conquer" strategy that applies off-the-shelf verifiers to sub-problems for which they perform better. While BaB can identify the sub-problems that are necessary to be split, it explores the space of these sub-problems in a naive "first-come-first-served" manner, thereby suffering from an issue of inefficiency to reach a verification conclusion. To bridge this gap, we introduce an order over different sub-problems produced by BaB, concerning with their different likelihoods of containing counterexamples. Based on this order, we propose a novel verification framework Oliva that explores the sub-problem space by prioritizing those sub-problems that are more likely to find counterexamples, in order to efficiently reach the conclusion of the verification. Even if no counterexample can be found in any sub-problem, it only changes the order of visiting different sub-problems and so will not lead to a performance degradation. Specifically, Oliva has two variants, including Oliva^GR, a greedy strategy that always prioritizes the sub-problems that are more likely to find counterexamples, and Oliva^SA, a balanced strategy inspired by simulated annealing that gradually shifts from exploration to exploitation to locate the globally optimal sub-problems. We experimentally evaluate the performance of Oliva on 690 verification problems spanning over 5 models with datasets MNIST and CIFAR-10. Compared to the state-of-the-art approaches, we demonstrate the speedup of Oliva for up to 25× in MNIST, and up to 80× in CIFAR-10.

Incremental verification is an emerging neural network verification approach that aims to accelerate the verification of a neural network N* by reusing the existing verification result (called a template ) of a similar neural network N . To date, the state‐of‐the‐art incremental verification approach leverages the problem splitting history produced by branch and bound ( BaB ) in verification of N , to select only a part of the sub‐problems for verification of N* , thus more efficient than verifying N* from scratch. While this approach identifies whether each sub‐problem should be re‐assessed, it neglects the information of how necessary each sub‐problem should be re‐assessed, in the sense that the sub‐problems that are more likely to contain counterexamples should be prioritized, in order to terminate the verification process as soon as a counterexample is detected. To bridge this gap, we first define a counterexample potentiality order over different sub‐problems based on the template, and then we propose Olive, an incremental verification approach that explores the sub‐problems of verifying N* orderly guided by counterexample potentiality. Specifically, Olive has two variants, including Olive g , a greedy strategy that always prefers to exploit the sub‐problems that are more likely to contain counterexamples, and Olive b , a balanced strategy that also explores the sub‐problems that are less likely, in case the template is not sufficiently precise. We experimentally evaluate the efficiency of Olive on 1445 verification problem instances derived from 15 neural networks spanning over two datasets MNIST and CIFAR‐10 . Our evaluation demonstrates significant performance advantages of Olive over state‐of‐the‐art classic verification and incremental approaches. In particular, Olive shows evident superiority on the problem instances that contain counterexamples, and performs as well as Ivan on the certified problem instances.

As Artificial Intelligence (AI) integrates into diverse areas, particularly in content generation, ensuring rightful ownership and ethical use becomes paramount, AI service providers are expected to prioritize responsibly sourcing training data and obtaining licenses from data owners. However, existing studies primarily center on safeguarding static copyrights, which simply treat metadata/datasets as non-fungible items with transferable/trading capabilities, neglecting the dynamic nature of training procedures that can shape an ongoing trajectory. In this paper, we presentIBis, a blockchain-based framework tailored for AI model training workflows. Our design can dynamically manage copyright compliance and data provenance in decentralized AI model training processes, ensuring that intellectual property rights are respected throughout iterative model enhancements and licensing updates. Technically,IBisintegrates on-chain registries for datasets, licenses and models, alongside off-chain signing services to facilitate collaboration among multiple participants. Further,IBisprovides APIs designed for seamless integration with existing contract management software, minimizing disruptions to established model training processes. We implementIBisusing Daml on the Canton blockchain. Evaluation results showcase the feasibility and scalability ofIBisacross varying numbers of users, datasets, models, and licenses.

Blockchains have been applied in different domains to guarantee data integrity and provide a decentralized computational infrastructure for executing smart contracts. Multiple blockchain-related patterns have been summarized by academics and industry practitioners covering different aspects, such as engineering applications on top of a blockchain, structuring smart contracts, and security. The existence of these patterns is both helpful and challenging for designers. Helpful, as the existence of these patterns means that developers do not need to recreate solutions to common problems. Challenging, as the multitude of patterns leaves a designer confused about when to adopt or adapt patterns. In this paper, we propose a decision model that assists developers and architects in selecting appropriate patterns for blockchain-based applications. The selection is based on the characteristics of the use cases and trade-offs implicit in the patterns. We evaluated the proposed decision model based on expert opinion regarding its correctness and usefulness in guiding the architecture design and understanding the rationale of various design decisions.

Although a massive amount of bandwidth is available at mm-waves, physics dictates the use of legacy frequencies in the sub 6-GHz range. This necessitates dynamic spectrum access in the face of exponentially growing spectral demands. However, disorganized spectrum sharing causes interference, leads to a chaotic situation, and loss of capacity. Moreover, it is difficult to ensure that the primary users are compensated for sharing their licensed bands. We propose a Blockchain-based platform to address these limitations. A digital token, called spectral token, is introduced to validate and track the use of a licensed frequency band while enforcing sequential access to spectrum by secondary users to avoid interference. The proposed platform enables both advertising and sensing based spectrum sharing under different leasing policies. Such sharing and leasing policies are coded into smart contracts, which digitally enforce the contractual clauses of the leasing agreement. When a deal is made, the smart contract automatically transfers the spectral token between primary and secondary users within the agreed time frame while paying the primary user in cryptocurrency. We developed a proof of concept solution using the Ethereum Blockchain to demonstrate the utility of the proposed platform and its throughput and latency characteristics.

Current cloud users pay for statically configured VM sizes irrespective of usage. It is more favorable for users to consume (and be billed for) just the right amount of resources necessary to satisfy the performance requirement of their applications. We take a novel perspective to enable such resource usage, where we assume that the cloud operator exposes a small, dynamic fraction of its infrastructure, corresponding resource specifications, and constraints to each application. We then propose a dynamic and computationally efficient reconfiguration scheme which comprises an Application Performance Model, a Cost Model, and a Reconfiguration algorithm. The performance model estimates application performance given specific resources. Cost model assigns a value to resource candidates made available to the application considering the lease expense, reconfiguration penalty, and operating income. A reconfiguration algorithm, assisted by the cost model, then makes optimal reconfiguration decisions. Simulation results for RUBiS and filebench-fileserver applications and Worldcup workloads show significant cost savings while meeting performance targets compared to rule-based scaling.

Elasticity is a key feature in Cloud Computing where virtualized resources are provisioned and de-provisioned via auto-scaling. However, auto-scaling in most Platform-as-a-Service (PaaS) systems is based on reactive, threshold-driven approaches. Such systems are incapable of catering to rapidly varying workloads, unless the associated thresholds are sufficiently low. Alternatively, maintaining low thresholds leads to resource over-provisioning under relatively stable workloads. Moreover, thresholds are not a good indication of QoS compliance, which is a key performance indicator of a cloud application. Hence, it is nontrivial to determine an optimum threshold while minimizing costs and meeting QoS demands. We propose inteliScaler, a proactive and cost-aware auto-scaling solution to address these issues by combining a predictive model, cost model, and a smart killing feature. An ensemble workload prediction mechanism is introduced based on time series and machine learning techniques for making accurate predictions on drastically different workload patterns. Utility of the solution is demonstrated using both simulations and empirical evaluations using Apache Stratos PaaS (deployed on the AWS EC2), as well as RUBiS and real-world workload traces. Results show significant QoS improvements and cost reductions by inteliScaler compared to a typical reactive and threshold-based PaaS auto-scaling solution.

Complex Event Processing (CEP) is a well-known technology in real-time Big Data processing systems. Performance of CEP engines is expected to scale with ever-increasing data rates and complex use cases. CEP operators like stream join and event patterns involve high computational complexity, hence, have a considerable impact on the overall query processing performance. Distributed event processing and CPU-level parallel event processing algorithms are common approaches for improving the performance. We explore how commodity massively parallel architectures like modern Graphics Processing Units (GPUs) can be utilized to improve the performance of frequently used CEP operators. We demonstrate how CEP operators such as event filter, event window, and stream join can be redesigned and implemented on GPUs to gain an order of magnitude improvement in throughput compared to a CPU-based implementation. This work is demonstrated using NVIDIA CUDA based implementation of CEP operators for Siddhi CEP engine on low-end GPUs. Moreover, this approach reduces event queuing at the incoming event queue, even with a large number of event streams, high arrival rates, and several complex queries. Consequently, the average latency experienced by incoming events is also reduced.

Collaborative peer-to-peer (P2P), grid, and cloud computing rely on resource discovery (RD) solutions to aggregate groups of multi-attribute, dynamic, and distributed resources. However, specific characteristics of real-world resources and queries, and their impact on P2P-based RD, are largely unknown. We analyze the characteristics of resources and queries using data from four real-world systems. These characteristics are then used to qualitatively and quantitatively evaluate the fundamental design choices for P2P-based multi-attribute RD. The datasets exhibit several noteworthy features that affect the performance. For example, compared to uniform queries, real-world queries are relatively easier to resolve using unstructured, superpeer, and single-attribute-dominated query-based structured P2P solutions, as queries mostly specify only a small subset of the available attributes and large ranges of attribute values. However, all the solutions are prone to significant load balancing issues, as the resources and queries are highly skewed and correlated. The implications of our findings for improving RD solutions are also discussed.

Large peer-to-peer systems exhibit the presence of communities based on user interests. Resources commonly shared within individual communities are in general relatively less popular and inconspicuous in the system-wide behavior. Hence, such communities are unable to benefit significantly from caching and replication that focus only on the most dominant queries. A community-based caching (CBC) solution that enhances both community-wide and system-wide lookup performance is proposed. CBC consists of a suboverlay formation scheme and a local-knowledge-based distributed caching (LKDC) algorithm. Suboverlays enable communities to forward queries through their members. While queries are forwarded, the LKDC algorithm causes members to identify and cache resources of interests to them, resulting in faster resolution of queries for popular resources within each community. Distributed local caching requires global information (e.g., hop count and popularity of contents) that is difficult and costly to obtain. However, by means of an analysis of globally optimal behavior and structural properties of the overlay, we developed the heuristic-based LKDC algorithm that not only relies on purely local information but also provides close-to-optimal caching performance. CBC is adaptive to changing popularity and user interests, works with any skewed distribution of queries, and introduces minimal modifications and overhead to the overlay network.

Emerging collaborative Peer-to-Peer (P2P) applications rely on resource discovery solutions to aggregate groups of heterogeneous, multi-attribute, and dynamic resources that are distributed. In the absence of data and understanding of real-life resource and query characteristics, design and evaluation of existing solutions have relied on many simplifying assumptions. We first present a summary of resource and query characteristics from PlanetLab. These characteristics are then used to evaluate fundamental design choices for multi-attribute resource discovery based on the cost of advertising/querying resources, index size, and load balancing. Simulation-based analysis indicates that the cost of advertising dynamic attributes is significant and in-creases with the number of attributes. Compared to uniform queries, real-world queries are relatively easier to resolve using unstructured, superpeer, and single-attribute dominated query based structured P2P solutions. However, they cause significant load balancing issues in all the designs where a few nodes are mainly involved in answering majority of queries and/or indexing resources. Moreover, cost of resource discovery in structured P2P systems is effectively O(N) as most range queries are less specific. Thus, many existing design choices are applicable only under specific conditions and their performances tend to degrade under realistic workloads.

Distributed, multi-attribute Resource Discovery (RD) is a fundamental requirement in collaborative Peer-to-Peer (P2P), grid, and cloud computing. We present an efficient and load balanced, P2P-based multi-attribute RD solution that consists of five heuristics, which can be executed independently and distributedly. First heuristic maintains a minimum number of nodes in a ring-like overlay consequently reducing the cost of resolving range queries. Second and third heuristics dynamically balance the key and query load by transferring keys to neighbors and by adding new neighbors when existing ones are insufficient. Last two heuristics, namely fragmentation and replication, form cliques of nodes that are placed orthogonal to the overlay ring to dynamically balance the highly skewed key and query loads while reducing the query cost. By applying these heuristics in the presented order, a RD solution that better responds to real-world resource and query characteristics is developed. Simulations using real workloads are used to demonstrate its efficacy.

Though resource discovery is a fundamental requirement in collaborative peer-to-peer, grid, and cloud computing, very little is known about resource/query characteristics and their impact on resource discovery. Fundamental design choices for distributed resource advertising and querying are evaluated in the context of existing practical systems. First, a generic model for cost of resource discovery is presented. Second, multi-attribute resource and query characteristics from Planet-Lab and SETI@home are presented. We observe that attributes of both resources and queries are highly skewed, correlated, queries are less specific, and Generalized Pareto distribution is suitable for capturing the distribution of most dynamic attributes and their rate of change. Based on these observations, different design choices are evaluated for resource discovery in terms of their cost of advertising/querying, latency, load balancing, and routing table size. The findings indicate that superpeer-based architectures have the potential to support large-scale resource aggregation as they simultaneously balance the cost and load.

Modeling and simulation of Peer-to-Peer (P2P) resources with correlated static and dynamic attributes is essential in application design, validation, and performance analysis. A novel mechanism is presented to generate realistic synthetic traces of multivariate static and dynamic attributes of P2P resources. The methodology is demonstrated using characteristics of PlanetLab node traces. First, a multi-attribute resource model is defined using a selected set of static and dynamic attributes. Second, characteristics of resources are presented. We observe that attribute values are correlated, follow a mixture of probability distributions, and time series of some of the dynamic attributes are nonstationary. Third, random vectors of static attributes are generated using empirical copulas that capture the entire dependence structure of multivariate distribution of attributes. Finally, time series of dynamic attributes are randomly drawn from a library of multivariate-time-series segments extracted from PlanetLab traces. These segments are identified by detecting the structural changes in time series corresponding to a selected attribute. Time series corresponding to rest of the attributes are split at the same breakpoints and randomly drawn together to preserve their contemporaneous correlation. Furthermore, a tool is developed to automate the synthetic data generation process and its output is validated using statistical tests.

Large Peer-to-Peer (P2P) systems for file transfer exhibit the presence of communities based on semantic, geographic, or organizational interests of users. Generally, resources commonly shared within individual communities are relatively unpopular and inconspicuous in the system-wide behavior. These communities are unable to benefit significantly from performance enhancement schemes such as caching that focus only on the most dominant queries. We propose a generic caching framework that enhances lookup performance of individual communities while providing even better performance to the dominant communities. The caching framework can be used with any structured P2P system that provides alternative paths to a given destination. Furthermore, the solution is adaptive to changing popularity and user interests, works with any skewed distribution of queries, needs small caches, utilizes local statistics, and introduces minimal modifications and overhead to the overlay network. Simulations based on Chord overlay show 40% reduction in average path length with individual communities indicating three times improvement in performance over system-wide caching.

Security is a prime concern in large-scale wireless sensor networks used for collaborative mission critical applications. A backbone network in the form of a cluster tree can enhance upper layer functions such as routing, broadcasting, in-network query processing and network management. A secure backbone based on the cluster tree enables secure upper layer functions and dynamic distribution of cryptographic keys among different nodes and users of collaborative networks. A secure cluster tree formation algorithm is presented that is independent of key pre-distribution scheme, network topology, and does not require a-priori neighborhood information or location awareness. Simulation based results show that the algorithm retains most of the desirable cluster and cluster tree characteristics while building the secure cluster tree. Availability of neighborhood information further improves the performance. Our simulations also suggest that hierarchical networks are more vulnerable to node capture than non-hierarchical networks.