In learning from demonstrations (LfD) for trajectory planning, end-to-end deep learning (DL) methods offer fast inference and adaptability to complex inputs. However, they are prone to cumulative errors due to limited expert time-series data, which poses challenges in safety-critical applications. To address this, we introduce bounded discontinuities in trajectory planning, with the bound adaptively determined via binary search. Two generative networks, trained in opposite directions, produce primitive trajectories. These are connected using the discontinuity-allowed multi-point RRT-connect (DAMP-RRT-connect) algorithm, which expands the trajectory while maintaining discontinuities within the bound. A sequence of navigation points directs the expansion. Experiments on aircraft landing and takeoff tasks at a non-towered airport demonstrate the robustness and efficiency of our approach. [Code]1

The virtual machine (VM) co-resident architecture of cloud computing enables simultaneous provision of multiple services to different users, but also makes these services vulnerable to co-resident attacks. For example, by establishing side channels, a malicious attacker can access and even corrupt services performed by other VMs co-residing on the same server as the attacker's VM (AVM). We model a threshold-voting-basedN-version programming service component with multiple independent versions simultaneously performing the same requested service to enhance the service reliability. However, the reliability enhancement can be greatly hindered by the co-resident attack, which may corrupt an adequate number of versions leading to a wrong output. We formulate and solve constrained optimization problems that determine the number of service component versions and the voting threshold to balance two conflicting service performance metrics: reliability (service component success probability) and vulnerability (service corruption attack success probability). Two cases respectively having certain and uncertain knowledge about the attacker's power in terms of the number of AVMs are considered. We also investigate impacts of different model parameters on the service performance as well as on solutions to the considered optimization problems through examples.

This article considers heterogeneous, cold standby systems performing missions with the fixed amount of work when a failure of an operating element results in a mission failure. A system is operating in a random environment modeled by the Poisson process of shocks. Each shock decreases the remaining lifetime of an operating element and, therefore, its preventive replacement (PR) is scheduled on experiencing the predetermined number of shocks. An important feature of the discussed model is that the failure can also occur during these PRs (task transfers) with two elements involved. The duration of the task transfer depends on the time from the start of a mission. The recursive equations for obtaining the mission success probability are derived and the corresponding numerical algorithm is developed. The number of shocks triggering elements’ replacements is obtained as a solution of the formulated optimization problem. The numerical example with the detailed analysis for a set of virtual machines operating in a cloud computing environment is presented.

Powered by virtualization, the cloud computing has brought good merits of cost effective and on-demand resource sharing among many users. On the other hand, cloud users face security risks from co-residence attacks when using this virtualized platform. Particularly, a malicious attacker may create side channels to steal data from a target user’s virtual machine (VM) that co-resides with the attacker’s VM on the same physical server. This article models a cloud service undergoing the co-residence data theft attacks. The threshold-voting-based${N}$-version programming (NVP) is implemented to improve the service reliability, where multiple service component versions (SCVs) are activated in parallel to perform the requested service. The final output is determined upon receiving a threshold number of identical outputs from the SCVs, immediately followed by canceling all outstanding SCVs to reduce expenses. Probabilistic models are first introduced to evaluate performance metrics of the considered service, including the data theft probability, service success probability, expected service operation time, and expected utility. Optimization problems are further solved to find the optimal number of SCVs maximizing the expected utility. Interactions among different model parameters and VM allocation policies, as well as their effects on the considered performance metrics and on the optimization solutions are studied through examples.

The real innovation of Internet of Things (IoT) can be spurred only when being combined with cloud computing, a paradigm that allows numerous users to simultaneously access configurable resources and services. However, serious vulnerability concerns have arisen from the virtual machine co-resident architecture of the IoT cloud. Specifically, co-resident attacks can be launched, where an attacker can access and corrupt a user's sensitive data/software by co-locating their virtual machines on the same physical server. Various solutions have been suggested in literature to mitigate negative effects of the co-resident attacks in the cloud environment. However, to the best of our knowledge no work has been performed for studying co-resident attacks in cloud systems withN-version programming (NVP), a popular redundancy technique for enhancing survivability of critical cloud service components. This paper makes original contributions by modeling IoT cloud system services implementing the NVP component redundancy, and evaluating the corruption probability of the NVP service component. Further, users’ policies on choosing the optimal number of service component versions are investigated through formulating and solving a new set of optimization problems with the objective to minimize the expected cost of losses of a cloud service provider. As demonstrated through examples, these policies can effectively help defend the NVP service component against the co-resident attacks in the cloud system.

In recent years, by using mobile cloud system (MCS) that integrates the cloud radio access network (C-RAN) with the mobile cloud computing (MCC) technology, mobile service provider (MSP) can efficiently enhance the capabilities of mobile users devices and handle the increasing mobile traffic to provide better quality of service (QoS). In a realistic scenario, the QoS of the MCS, i.e., service performance are extremely affected by resources allocation (such as the number of Virtual Machine (VM) and the CPU power of each VM). Therefore, the service performance evaluating of MCS is very necessary, which can benefit for optimizing resource allocation and task scheduling strategy. However, the studies of the performance evaluation for the MCS are still lacking in previous works. In this paper, we propose a performance modeling approach to evaluate the service performance of the MCS. A multi-queue model is first used to model the request execution process of the MCS. To the best as we know, we are the first one who uses the multi-queue model modeling the MCS. In the experiment part, two simulations are designed to verify this modeling approach. Then, we change the value of each performance-related factors to reveal their impact on the service performance of the MCS. The results show that when resources are lacking or just enough in the MCS, increasing the computing resources and the available machine in the C-RAN or the MCC is the fastest way to improve the service performance of the MCS.

The openness and inclusiveness of Internet of Things (IoT) inevitably result in security risks. Privacy protection in data publishing in IoT has become one of the most important topics. For data owners, besides the privacy of released information, they are increasingly concerned about the privacy of unreleased data because data users are particularly keen to analyze results from the released information. Our experimental results demonstrate that a small number of known data could be successfully utilized to estimate the unknowns. To address the unreleased data privacy while guaranteeing the utility of released data, we propose a privacy-aware controllable compressed data publishing strategy that could resist sparse estimation attacks efficiently. Specifically, by introducing compressive sensing (CS) technology to achieve compressed data publishing, communication overhead is reduced greatly. With the help of CS reconstruction error, the privacy protection capability of unreleased data is enhanced under the premise of the utility of released data. What is more, users are divided into the data user and the authorized user to realize controllability of the data owner. For the same compression data, the data user with general permissions is limited to the released data from the data owner; while the authorized user is entitled to access the whole data. Furthermore, we analyze the upper bound of the released data number that could satisfy the privacy requirement of the unreleased data. The experimental results demonstrate that the utility of the released data, the privacy of the unreleased data, and the compressibility all achieve desirable effect.

Energy-efficient cloud computing has recently attracted much attention, where not only performance but also energy consumption are important metrics to be considered for designing rational resource scheduling strategies. Most of existing approaches for achieving energy efficient computing focus on connecting these two metrics and balancing the tradeoff between them, which however is inadequate because another important factor reliability is not considered. In fact, both virtual machine (VM) failures and server failures inevitably interrupt execution of a cloud service, and eventually result in spending more time and consuming more energy on completing the cloud service. Therefore, reliability significantly affects service performance and energy consumption, and thus they should not be handled separately. Connecting these correlated metrics is essential for making more precise evaluation and further for developing rational cloud resource scheduling strategies. In this paper, we present a correlated modeling approach applying Semi-Markov models, the Laplace-Stieltjes transform (LST), a Bayesian approach to analyze reliability-performance (R-P) and reliability-energy (R-E) correlations for cloud services using a retrying fault recovery mechanism. A recursive method is also proposed for modeling the correlations for cloud services using a check-pointing fault recovery mechanism. The proposed correlation models can be used to calculate the expected service time and energy consumption for completing a cloud service. Moreover, the models can contribute to analyzing the expected performance-energy tradeoff. We formulate the expected performance-energy optimization problem by describing performance and energy consumption metrics as functions of assigned CPU frequencies. Finally, we use a derivation approach to determine Pareto optimal solutions for the formulated optimization problem. Illustrative examples are provided.

Multifactorial evolutionary algorithm (MFEA) exploits the parallelism of population-based evolutionaryalgorithm and provides an efficient way to evolve individuals for solving multiple tasks concurrently.Its efficiency is derived by implicitly transferring the genetic information among tasks.However, MFEA doesn?t distinguish the information quality in the transfer compromising the algorithmperformance. We propose a group-based MFEA that groups tasks of similar types and selectivelytransfers the genetic information only within the groups. We also develop a new selection criterionand an additional mating selection mechanism in order to strengthen the effectiveness andefficiency of the improved MFEA. We conduct the experiments in both the cross-domain and intra-domainproblems.

Meme automaton is an adaptive entity that autonomously acquires an increasing level of capability and intelligence through embedded memes evolving independently or via social interactions. This paper begins a study on memetic multiagent system (MeMAS) toward human-like social agents with memetic automaton. We introduce a potentially rich meme-inspired design and operational model, with Darwin's theory of natural selection and Dawkins' notion of a meme as the principal driving forces behind interactions among agents, whereby memes form the fundamental building blocks of the agents' mind universe. To improve the efficiency and scalability of MeMAS, we propose memetic agents with structured memes in this paper. Particularly, we focus on meme selection design where the commonly used elitist strategy is further improved by assimilating the notion of like-attracts-like in the human learning. We conduct experimental study on multiple problem domains and show the performance of the proposed MeMAS on human-like social behavior.

Cloud computing is a new emerging technology aimed at large-scale resource sharing and service-oriented computing. To achieve the efficient use of cloud resources for supporting a cloud service, many important factors need to be considered, particularly, reliability, performance, and power consumption of the cloud service. Evaluation of these metrics is essential for further designing rational resource scheduling strategies. However, these metrics are closely related; they do affect one another. The cloud system should consider correlations among the metrics to make more precise evaluation. Most of the existing approaches and models handle these metrics separately, and thus they cannot be used to study the correlations. This paper presents a new hierarchical correlation model for analyzing and evaluating these correlated metrics, which encompasses Markov models, queuing theory, and a Bayesian approach. Various distinctive characteristics of the cloud system are investigated and captured in the model, such as multiple virtual machines (VMs) hosted on the same server, common cause failures of co-located VMs caused by server failures, and logical mapping mechanisms for multicore CPUs. Moreover, for evaluating and balancing the tradeoff between performance and power consumption, a tradeoff parameter and a pure profit optimization model are developed based on the presented correlation model. Numerical examples are provided.

Point-of-interest(POI) recommendation becomes a valuable service in location-based social networks. Based on the norm that similar users are likely to have similar preference of POIs, the current recommendation techniques mainly focus on users' preference to provide accurate recommendation results. This tends to generate a list of homogeneous POIs that are clustered into a narrow band of location categories(like food, museum, etc.) in a city. However, users are more interested to taste a wide range of flavors that are exposed in a global set of location categories in the city.In this paper, we formulate a new POI recommendation problem, namely top-K location category based POI recommendation, by introducing information coverage to encode the location categories of POIs in a city.The problem is NP-hard. We develop a greedy algorithm and further optimization to solve this challenging problem. The experimental results on two real-world datasets demonstrate the utility of new POI recommendations and the superior performance of the proposed algorithms.

Software as a Service is the most adopted cloud service (46%) compared with Infrastructure as a Service (IaaS) (35%) and Platform as a Service (PaaS) (34%) [1]. Currently, the capability of discovering a SaaS of interest online across multiple cloud providers and reviews websites is a significant challenge, especially when using general search mechanisms (Google and Yahoo!) and search tools provided by existing reviews and directories. Discovering a SaaS is time-consuming, requiring consumers to browse several websites to select the appropriate service. This paper addresses the issues related to the efficient discovery of SaaS across review websites by developing the SaaS Nutch Hadoop-based Crawler Engine - SaaS Nhbased Crawler. The crawler is capable of crawling cloud reviews to find SaaSs of interest and enable the establishment of a central repository that could be used to discover SaaSs much more efficiently. The results show that the SaaS Nhbased crawler can effectively crawl review websites and provide a list of the latest SaaS being offered.

Cloud computing has recently emerged as an important filed with numerous novel features, particularly, large-scale resource integration and virtualized resource provisioning. Since a cloud system essentially aims at service-oriented computing, service performance becomes the primary metric that needs analyzing in detail. However, in a realistic scenario, operation of virtual machines (VM) may be interrupted by random resource failures. This demonstrates that service performance is indeed affected by resource reliability. Thus, connecting performance and reliability is essential for making more precise evaluation. In this paper, we present a theoretical modeling approach for performability analysis of cloud services and the cloud system. This flexible modeling approach first builds two tractable submodels that consider an important correlation factor (i.e., available resource capacity that is not only decided by reliability but also has a significant effect on performance) to ensure the required fidelity. Then, a Bayesian method is applied to connect the submodels, which can make our performability model more scalable. In contrast to a monolithic modeling method, our approach that combines interacting submodels can effectively reduce computing complexity for a large-scale cloud system. Numerical examples are illustrated.

Adverse drug reaction (ADR) is a common clinical problem, sometimes accompanying with high risk of mortality and morbidity. It is also one of the major factors that lead to failure in new drug development. Unfortunately, most of current experimental and computational methods are unable to evaluate clinical safety of drug candidates in early drug discovery stage due to the very limited knowledge of molecular mechanisms underlying ADRs. Therefore, in this study, we proposed a novel na€ıve Bayesian model for rapid assessment of clinical ADRs with frequency estimation. This model was constructed on a gene-ADR association network, which covered 611 US FDA approved drugs, 14,251 genes, and 1,254 distinct ADR terms. An average detection rate of 99.86 and 99.73 percent were achieved eventually in identification of known ADRs in internal test data set and external case analyses respectively. Moreover, a comparative analysis between the estimated frequencies of ADRs and their observed frequencies was undertaken. It is observed that these two frequencies have the similar distribution trend. These results suggest that the naıve Bayesian model based on gene-ADR association network can serve as an efficient and economic tool in rapid ADRs assessment.

Influence maximization problem is to find a set of seed nodes in a social network such that their influence spread is maximized under certain propagation models. A few algorithms have been proposed for solving this problem. However, they have not considered the impact of novelty decay on influence propagation, i.e., repeated exposures will have diminishing influence on users. In this paper, we consider the problem of influence maximization with novelty decay (IMND). We investigate the effect of novelty decay on influence propagation on real-life datasets and formulate the IMND problem. We further analyze the problem properties and propose an influence estimation technique. We demonstrate the performance of our algorithms on four social networks.

Memetic multiagent system emerges as an enhanced version of multiagent systems with the implementation of meme-inspired computational agents. It aims to evolve human-like behavior of multiple agents by exploiting the Dawkins' notion of a meme and Universal Darwinism. Previous research has developed a computational framework in which a series of memetic operations have been designed for implementing human-like agents. This paper will focus on improving the human-like behavior of multiple agents when they are engaged in social interactions. The improvement is mainly on how an agent shall learn from others and adapt its behavior in a complex dynamic environment. In particular, we design a new mechanism that supervises how the agent shall select one of the other agents for the learning purpose. The selection is a trade-off between the elitist and like-attracts-like principles. We demonstrate the desirable interactions of multiple agents in two problem domains.

This paper considers a system consisting ofnlinearly ordered multistate components. Each component can have different states: from complete failure, up to perfect functioning. A performance rate is associated with each state. The system fails if in each of at leastmconsecutive overlapping groups ofrconsecutive components (windows) the sum of the performance rates of components belonging to the group is lower than a minimum allowable level. It is shown that, in the case of different components, the system reliability depends on their arrangement. The optimal arrangement problem is formulated, and a numerical tool for solving this problem is suggested. The tool uses an extended universal moment generating function technique for system reliability evaluation, and a genetic algorithm for optimization. Examples of system reliability optimization are presented.

This paper proposes a new model that generalizes the linear multi-state sliding window system. In this model, the system consists ofnlinearly ordered multi-state elements. Each element can have different states spanning from complete failure up to perfectly functioning. A performance rate is associated with each state. The system fails if the gap between any pair of groups ofrconsecutive elements having the cumulative performance lower than a minimum allowable levelWis less thanmgroups ofrconsecutive elements. An algorithm for system reliability evaluation is suggested which is based on an extended universal moment generating function. Examples of evaluating system reliability and elements' reliability importance indices are presented.

This paper generalizes the linear multistate consecutively-connected system model by introducing allowable gaps. The new model consists of N +1 linearly ordered nodes. Some of these nodes contain statistically independent multistate elements with different characteristics. Each element j can provide a connection between the node to which it belongs and Xjnext nodes, where Xjis a discrete random variable with known probability mass function. The system fails if it contains at least m consecutive nodes not connected with any previous node (m consecutive gaps). An algorithm based on the universal generating function method is suggested for the system reliability evaluation. Illustrative examples are presented.

Computing software and systems have become increasingly large and complex. As their dependability and autonomy are of great concern, self-healing is an ongoing challenge. This paper presents an innovative model and technology to realize the self-healing function under the real-time requirement. The proposed approach, different from existing technologies, is based on a new concept defined as consequence-oriented diagnosis and healing. Derived from the new concept, a prototype model for proactive self-healing actions is presented. Then, a hybrid diagnosis tool is proposed that takes advantages from the Multivariate Decision Diagram, Fuzzy Logic, and Neural Networks, achieving an efficient, effective, accurate, and intelligent result. The consequence-oriented diagnosis and self-healing function is also implemented. The experimental results exhibit that the innovative system is very effective and precise in predicting the consequence, and in preventing resulting software and system failures.

Bayesian networks are known for providing an intuitive and compact representation of probabilistic information and allowing the creation of models over a large and complex domain. Bayesian learning and reasoning are nontrivial for a large Bayesian network. In parallel, it is a tough job for users (domain experts) to extract accurate information from a large Bayesian network due to dimensional difficulty. We define a formulation of local components and propose a clustering algorithm to learn such local components given complete data. The algorithm groups together most inter-relevant attributes in a domain. We evaluate its performance on three benchmark Bayesian networks and provide results in support. We further show that the learned components may represent local knowledge more precisely in comparison to the full Bayesian networks when working with a small amount of data.

As Bayesian networks become widely accepted as a normative formalism for diagnosis based on probabilistic knowledge, they are applied to increasingly larger problem domains. These large projects demand a systematic approach to handle the complexity in knowledge engineering. The needs include modularity in representation, distribution in computation, as well as coherence in inference. Multiply Sectioned Bayesian Networks (MSBNs) provide a distributed multiagent framework to address these needs. According to the framework, a large system is partitioned into subsystems and represented as a set of related Bayesian subnets. To ensure exact inference, the partition of a large system into subsystems and the representation of subsystems must follow a set of technical constraints. How to satisfy these goals for a given system may not be obvious to a practitioner. In this paper, we address three practical modeling issues.