Activities of Daily Living Scale (ADLs) is widely used to evaluate living abilities of the patients and the elderly. Most of the currently proposed approaches for tracking indicators of ADLs are human-centric. Considering the privacy concerns of the human-centric approaches, a new thing-centric sensing system, named TaRad, for detecting some indicators of ADLs (i.e. using fridge, making a phone call), through identifying vibration of objects when a person interacts with objects. It consists of action transceivers (named ViNode), smart phones and a server. By taking into account the limited computation resource of the action transceiver, and the drift and accuracy issues of the cheap sensor, a method of extracting features from the vibration signal, named ViFE, along with a light-weight activity recognition method, named ViAR, have been implemented in ViNode. Besides, an operator recognition method, named ViOR, has been proposed to recognize the acting person who generates vibration of action transceiver, when two or more people exist simultaneously within an area. Experimental results verify the performance of TaRad with different persons, in terms of the sensitivity to correctly detect the activities, and probability to successfully recognize the operators of the activities.
The wide-spread availability of open WiFi networks on smart cities can be considered an advanced service for citizens. However, a device connecting to WiFi network access points gives away its location. On the one hand, the access point provider could collect and analyse the ids of connecting devices, and people choose whether to connect depending on the degree of trust to the provider. On the other hand, an app running on the device could sense the presence of nearby WiFi networks, and this could have some consequences on user privacy. Based on permission levels and mechanisms proper of Android OS, this paper proposes an approach whereby an app attempting to connect to WiFi networks could reveal to a third part the presence of some known networks, thus a surrogate for the geographical location of the user, while she is unaware of it. This is achieved without resorting to GPS readings, hence without needing dangerous-level permissions. We propose a way to counteract such a weakness in order to protect user privacy.
In this paper, we describe a hybrid MPI implementation of a discontinuous Galerkin scheme in Computational Fluid Dynamics which can utilize all the available processing units (CPU cores or GPU devices) on each computational node. We describe the optimization techniques used in our GPU implementation making it up to 74.88x faster than the single core CPU implementation in our machine environment. We also perform experiments on work partitioning between heterogeneous devices to measure the ideal load balance achieving the optimal performance in a single node consisting of heterogeneous processing units. The key problem is that CFD workloads need to allocate large amounts of both host and GPU device memory in order to compute accurate results. There exists an economic burden, not to mention additional communication overheads of simply scaling out by adding more nodes with high-end scientific GPU devices. In a micro-management perspective, workload size in each single node is also limited by its attached GPU memory capacity. To overcome this, we use ZFP, a floating-point compression algorithm to save at least 25
An increasing number of organisations are harnessing the benefits of hybrid cloud adoption to support their business goals and achieving privacy and control in a private cloud whilst enjoying the on-demand scalability of the public cloud. However the complexity introduced by the combination of the public and private clouds worsens visibility in cloud monitoring with regards compliance to given business constraints. Load balancing as a technique for evenly distributing workloads can be leveraged together with processing mining to help ease the monitoring challenge. In this paper we propose a load balancing approach to distribute workloads in order to minimise violations to specified business constraints. The scenario of a hospital consultation process is employed as a use case in monitoring and controlling Octavia load balancing-as-a-service in OpenStack. The results show a co-occurrence of constraint violations and Octavia L7 Policy creation, indicating a successful application of process mining monitoring in load balancing.
Coordination of Cyberphysical Systems is an increasingly relevant concern for distributed systems engineering, mostly due to the rise of the Internet of Things vision in many application domains. Against this background, Speaking Objects has been proposed as a vision of future smart objects coordinating their collective perception and action through argumentation. Along this line, in this paper we describe a Proof-of-Concept implementation of the Speaking Objects vision in a smart home deployment.
Decentralized Online Social Networks (DOSNs) have been proposed as an alternative solution to the current centralized Online Social Networks (OSNs). Online Social Networks are based on centralized architecture (e.g., Facebook, Twitter, or Google+), while DOSNs do not have a service provider that acts as central authority and users have more control over their information. Several DOSNs have been proposed during the last years. However, the decentralization of the OSN requires efficient solutions for protecting the privacy of users, and to evaluate the trust between users. Blockchain represents a disruptive technology which has been applied to several fields, among these also to Social Networks. In this paper, we propose a manageable, user-driven and auditable access control framework for DOSNs using blockchain technology. In the proposed approach, the blockchain is used as a support for the definition of privacy policies. The resource owner uses the public key of the subject to define flexible role-based access control policies, while the private key associated with the subjects Ethereum account is used to decrypt the private data once access permission is validated on the blockchain. We evaluate our solution by exploiting the Rinkeby Ethereum testnet to deploy the smart contract, and to evaluate its performance. Experimental results show the feasibility of the proposed scheme in achieving auditable and user-driven access control via smart contract deployed on the Blockchain.
Table tennis stroke recognition is very important for athletes to analyze their sports skills. It can help players to regulate hitting movement and calculate sports consumption. Different players have different stroke motions, which makes stroke recognition more difficult. In order to accurately distinguish the stroke movement, this paper uses body sensor networks (BSN) to collect motion data. Sensors collecting acceleration and angular velocity information are placed on the upper arm, lower arm and back respectively. Principal component analysis (PCA) is employed to reduce the feature dimensions and support vector machine (SVM) is used to recognize strokes. Compared with other classification algorithms, the final experimental results (97.41% accuracy) illustrate that the algorithm proposed in the paper is effective and useful.
Fog computing provides an efficient solution for mobile computing offloading, keeping tight constraints on the response time for real-time applications. The paper takes into account the variation of tasks by introducing the joint distribution function of the required processing volume and data size to be transmitted. We propose an offloading criterion based on processing and data volumes of tasks and develop an analytical framework for the evaluation of the average response time and average energy consumption of mobile devices. The developed framework is used in the case study.
The availability of distributed renewable energy sources (RES), such as photo-voltaic panels, allows to locally consume or accumulate energy, avoiding power peaks and loss along the power network. However, as the number of utilities in a household or a building increases, and the energy must be equally and intelligently shared among the utilities and devices, demand side management systems must exploit new solution for allowing such energy usage optimisation. The current trends of demand side management systems highly exploit loads shifting, as a concrete solution to align consumption to the fluctuating produced power, and to maximise the energy utilisation avoiding its wastage. Moreover the introduction, in the latest years, of e-cars has given a boost to smart charging, as it can increase the flexibility that is necessary for maximising the self-consumption. However we strongly believe that a performing demand side management system must be able to learn and predict user’s habits and energy requirements of her e-car, to better schedule the loads shifting and reduce energy wastage. This paper focuses on the e-car utilisation, investigating the exploitation of machine learning techniques to extract and use such knowledge from the power measures at charging plug.
Users want websites to deliver rich content quickly. However, rich content often comes from separate subdomains and requires additional DNS lookups, which negatively impact web performance metrics such as First Meaningful Paint Time, Page Load Time, and the Speed Index. In this paper we investigate the impact of DNS lookups on web performance and propose Multi-Resolution DNS (MR-DNS) to reduce DNS resolutions through response batching. Our results show that MR-DNS has the potential to improve Page Load Time around 14
An increasing number of enterprises deploy their business applications in green data centers (GDCs) to address irregular and drastic natures in task arrival of global users. GDCs aim to schedule tasks in the most cost-effective way, and achieve the profit maximization by increasing green energy usage and reducing brown one. However, prices of power grid, revenue, solar and wind energy vary dynamically within tasks’ delay constraints, and this brings a high challenge to maximize the profit of GDCs such that their delay constraints are strictly met. Different from existing studies, a Temporal-variation-aware Profit-maximized Task Scheduling (TPTS) algorithm is proposed to consider dynamic differences, and intelligently schedule all tasks to GDCs within their delay constraints. In each interval, TPTS solves a constrained profit maximization problem by a novel Simulated-annealing-based Chaotic Particle swarm optimization (SCP). Compared to several state-of-the-art scheduling algorithms, TPTS significantly increases throughput and profit while strictly meeting tasks’ delay constraints.
Apps running on a smartphone have the possibility to gather data that can act as a fingerprint for their user. Such data comprise the ids of nearby WiFi networks, features of the device, etc., and they can be a precious asset for offering e.g. customised transportation means, news and ads, etc. Additionally, since WiFi network ids can be easily associated to GPS coordinates, from the users frequent locations it is possible to guess their home address, their shopping preferences, etc. Unfortunately, existing privacy protection mechanisms and permissions on Android OS do not suffice in preventing apps from gathering such data, which can be considered sensitive and not to be disclosed to a third part. This paper shows how an app using only the permission to access WiFi networks could send some private data unknowingly from the user. Moreover, an advanced mechanism is proposed to shield user private data, and to selectively obscure data an app could spy.
Foraging constitutes one of the main benchmarks in robotic problems. It is known as the act of searching for objects/tokens and, when found, transport them to one or multiple locations. Swarm intelligence based algorithms have been widely used in foraging problem. The ambient light sensors technology in nowadays robots makes easy using and implementing luminous swarm intelligence-based algorithms such as the Firefly and the Glow-worm algorithms. In this paper, we propose a swarm intelligence-based foraging algorithm called Lévy walk and Firefly Foraging Algorithm (LFFA) which is a hybridizing of the two algorithms Lévy Walk and Firefly Algorithm. Numerical experiments to test the performances are conducted on the ARGoS robotic simulator.
Internet of Things (IoT) sensors generate massive streaming data which needs to be processed in real-time for many applications. Anomaly detection is one popular way to process such data and discover nuggets of information. Various machine learning techniques for anomaly detection rely on pre-labelled data which is very expensive and not feasible for streaming scenarios. Autoencoders have been found effective for unsupervised outlier removal because of their inherent ability to better reconstruct data with higher density. Our work aims to leverage this principle to investigate approaches through which the optimal threshold for anomaly detection can be obtained in an automated and adaptive fashion for streaming scenarios. Rather than experimentally setting an optimal threshold through trial and error, we obtain the threshold from the reconstruction errors of the training data. Inspired by image processing, we investigate how thresholds set by various statistical approaches can perform in an image dataset.
In the last decade, GPGPU virtualization and remoting have been among the most important research topics in the field of computer science and engineering due to the rising of cloud computing technologies. Public, private, and hybrid infrastructures need such virtualization tools in order to multiplex and better organize the computing resources. With the advent of novel technologies and paradigms, such as edge computing, code offloading in mobile clouds, deep learning techniques, etc., the need for computing power, especially of specialized hardware such as GPUs, has skyrocketed. Although many GPGPU virtualization tools are available nowadays, in this paper we focus on improving GVirtuS, our solution for GPU virtualization. The contributions in this work focus on the CUDA plug-in, in order to provide updated performance enabling the next generation of GPGPU code offloading applications. Moreover, we present a new GVirtuS implementation characterized by a highly modular approach with a full multithread support. We evaluate and discuss the benchmarks of the new implementation comparing and contrasting the results with the pure CUDA and with the previous version of GVirtuS. The new GVirtuS yielded better results when compared with its previous implementation, closing the gap with the pure CUDA performance and trailblazing the path for the next future improvements.
Neural network architectures have demonstrated to achieve impressive results across a wide range of different domains. The availability of very large datasets makes possible to overcome the limitation of the training stage thus achieving significant level of performance. On the other hand, even though the advancements in GPU hardware, training a complex neural network model still represents a challenge. Long time is required when the computation is demanded to a single machine. In this work, a distributed training approach for 3DPyraNet model built for a specific domain, that is the emotion recognition from videos, is discussed. The proposed work aims at distributing the training procedures over the nodes of the Intel DevCloud Platform and demonstrating how the training performance are affected in terms of both computational demand and achieved accuracy compared to the use of a single machine. The results obtained in an experimental design suggests the feasibility of the approach for challenging computer vision tasks even in presence of limited computing power based on exclusive use of CPUs.
The Mediterranean area is subject to a range of destructive weather events, including middle-latitudes storms, Mediterranean sub-tropical hurricane-like storms (“medicanes”), and small-scale but violent local storms. Although predicting large-scale atmosphere disturbances is a common activity in numerical weather prediction, the tasks of recognizing, identifying, and tracing trajectories of such extreme weather events within weather model outputs remains challenging. We present here a new approach to this problem, called StormSeeker, that uses machine learning techniques to recognize, classify, and trace the trajectories of severe storms in atmospheric model data. We report encouraging results detecting weather hazards in a heavy middle-latitude storm that struck the Ligurian coast in October 2018, causing disastrous damages to public infrastructure and private property.
The Entropy has been used to characterize the neighbourhood of a sample on the base of its k Nearest Neighbour when data are imbalanced and many measures of Entropy have been proposed in the literature to better cope with vagueness, exploiting fuzzy logic, rough set theory and their derivatives. In this paper, a rough extension of Entropy is proposed to measure uncertainty and ambiguity in the neighbourhood of a sample, using the lower and upper approximations from rough–fuzzy set theory in order to compute the Entropy of the set of the k Nearest Neighbours of a sample. The proposed measure shows better robustness to noise and allows a more flexible modeling of vagueness with respect to the Fuzzy Entropy.
In recent years the biological data, represented for computational analysis, has increased in size terms. Despite the representation of the latter is demanded to specific file format, the analysis and managing overcame always more difficult due to high dimension of data. For these reasons, in recent years, a new computational framework, called Hadoop for manage and compute this data have been introduced. Hadoop is based on MapReduce paradigm to manage data in distributed systems. Despite the gain of performance obtained from this framework, our aim is to introduce a new compression method DSRC by decreasing the size of output file and make easy its processing from ad-hoc software. Performance analysis will show the reliability and efficiency achieved by our implementation.
In this paper we consider a CoT (Cloud of Things) scenario where agents cooperate to perform complex tasks. Agents have to select reliable partners and, in some cases, they don’t have enough information about their peers. In order to support agents in their choice and to maximize the benefits during their cooperation, we combined several contributions. First of all, we designed a trust model which exploits the recommendations coming from the ego networks of the agents. Secondly, we propose to partition the agents in groups by exploiting trust relationships to allow agents to interact with the most reliable partners. To this aim, we designed an algorithm named DAGA (Distributed Agent Grouping Algorithm) to form agent groups by exploiting available reliability and reputation and the results obtained in a simulated scenario confirmed its potential advantages.
