International Journal of High Performance Systems Architecture (11 papers in press)
An amelioration of the Skyline algorithm used in the Cloud Service Research and Selection System
by Manar ABOUREZQ, Abdellah IDRISSI, Hajar REHIOUI
Abstract: Cloud computing is both a technological and a business model revolution that capitalized on many long proven technologies. Its ease of use resulted in its wide adoption, and users are increasingly faced with a large choice of cloud services. In this context, we have developed the Cloud Service Research and Selection System (CSRSS) that allows users to specify the requirements of the cloud services they want and select those that best meet these requirements. The CSRSS uses an algorithm based on the Skyline and gave some encouraging firs results. In this paper, we explore two approaches to improve the performance of the used algorithm and decrease its time complexity by reducing the number of I/O operations necessary to compute the Skyline
Keywords: Cloud Computing; Cloud services; Skyline algorithm; Block-Nested Loops algorithm.
HYBRID ZIGBEE RFID MODEL FOR TAG DETECTION AND ENERGY TRADEOFF
by S. Bagirathi, Sharmila Sankar, M. Sandhya
Abstract: Active RFID systems are utilized widely in extensive applications and have limited radio communication. Collisions subsist during the tag detection process of the reader and surmounting these collisions and incrementing tag identification rate are the challenges in RFID system. Active RFID tags are battery powered and the occurrence of collisions during tag detection causes lassitude of tags battery life with the decrementation in system throughput, increase of execution time and communication overhead. Zigbee network provides multi hop data transmission and covers long distances. Effective tag identification with remarkable high throughput remains as a key challenge in RFID based applications. In this paper we propose a hybrid Zigbee RFID system model implemented with Dynamic Frame Slotted Grouping Collision Resolution (DFSGCR) protocol. The tags in the system are grouped randomly using regression clustering technique for inter and intra group collision minimization. Unique Group IDs are assigned to each group in the network. Reader uses the Group ID as an additional parameter along with frame size and random number during tag detection process. During every iteration of tag estimation the frame size is adjusted predicated on the comparison of total number of singleton and idle slots obtained from precedent frame slot structure against the number of collision slots without eliminating idle slots. Simulations were performed utilizing Opnet simulator which showed decreased collision rate, system overhead with amended performance in the metrics of execution time, tag identification rate, energy efficiency and a maximum throughput of 0.63 was achieved when compared to other tag detection methods.
Keywords: RFID; Zigbee; Tag collision; Dynamic frame slotted aloha; Collision rate; Energy efficiency; Opnet simulator.
Special Issue on: Recent Advances in the Security of Multimedia Big Data in Semantic Web-based Social Networks
A new correntropy-based level set algorithm using local robust statistics information
by Sheng Wang, Xiaoliang Jiang
Abstract: Intensity inhomogeneity or noise usually appear in various kinds of images, which cause a challenging task in image segmentation. To solve these issues, a novel correntropy-based level set algorithm utilizing local robust statistics information is introduced. In the proposed method, the modified local image fitting (MLIF) equation is built by describing the difference between the images of fitted and local robust statistics. Then, by using the correntropy criterion, the MLIF model can automatically emphasize the weight coefficient of the samples that are approximately to the gray means. In this case, the new guided energy term can accurately process images with weak edge and more adaptive to noise. Finally, we introduce a level set regularization terms to remove re-initialization process. Experiments on a lot of images demonstrate our method has good segmentation ability on the part of visual perception and robustness, as compared with traditional algorithms.
Keywords: Correntropy-based; level set; local robust statistics.
A Video Image Detection Approach Based on Cooperative Positioning
by Cai Aiping
Abstract: Target detection tasks usually need to be performed with the help of cooperative positioning in machine vision positioning systems with multiple viewing angles and cameras. For traditional target positioning tasks, classifiers are used to distinguish foregrounds and backgrounds, or new dictionaries of visions are created by learning features of samples. Models built by classifier algorithms in most cases are complicated, compromising algorithms efficiency while ensuring the accuracy of the models. A detection approach based on cooperative positioning of locations and mutex constraints is proposed in this paper, and with this approach, the efficiency of classification can be improved. First, a mutex matrix of consecutive frame images in video clips is worked out, and a target function model is built by performing exclusive or operation on this matrix and the original mutex matrix. Then L-K optical flow approach is used to get an optical flow map and the motion-related priori information of the candidate objects. Then a mutex matrix based on overlapping degree and aspect ratio is used to constrain candidate objects overlapped and greatly different in appearance from being selected from single images. When either mutex is satisfied, the two kinds of candidate targets cannot be included in the maximum weight clique simultaneously, and thereby the problem of cooperative positioning modeling is solved. According to the test results, better detection effects can be generated since the location and scale of one object in one video clip will not undergo any sudden and drastic change.
Keywords: image classification;co-localization;video detection; mutex constrained.
Applications of Robust Recognition Technology in the Foreign Language Speech Assessment System
by Weibo Huang, Jiejing Lu, Jianghui Liu
Abstract: This study proposes an improved speech assessment algorithm and an improved English (as a foreign language in China) speech assessment system based on robust speech recognition in order to evaluate learners' English pronunciation more accurately in noise environment. The traditional assessment framework is studied by the literature analysis method and survey research. The principle of speech recognition and robust speech recognition are further studied. At the same time, the speech assessment algorithm and the system based on the feature compensation algorithm of Vector Taylor Series (VTS) is developed. Finally, the developed system is improved and analyzed by Perceptual Assessment of Speech Quality (PESQ) and Mean Opinion Score (MOS). The results show that the MOS of the improved English (as the foreign language) speech assessment system is higher and the recognition performance is better than that of the traditional system in the noise environment.
Keywords: English Speech Assessment System; Noise Conditions; Robust Speech Recognition; Feature Compensation.
Facial Beauty Prediction via Deep Cascaded Forest
by YiKui Zhai, Peilun Lv, Wenbo Deng, Xueyin Xie, Cuilin Yu, Junying Gan, Junying Zeng, Zilu Ying, Ruggero Donida Labati, Vincenzo Piuri, Fabio Scotti
Abstract: Face Beauty Prediction (FBP), which is a prediction based on the classification of human facial beauty, has been applied in some social platforms and entertainment software. However, among the various approaches to FBP, methods based convolutional network is too complicated, and traditional methods cannot achieve the desired performance. In this paper, we propose a method for FBP via deep cascade forest. This method uses multi-grained scanning to obtain the features of the image, and uses multiple random forests to enhance the features. Then multiple classifiers to form a new classifier, which is used for predicting the acquired features to complete the FBP task. This method shows the advantages of feature extraction and relatively high prediction accuracy in 10,000 facial beauty datasets (10TFBD). And we are optimized for the cascade forest part and further improved the prediction accuracy. Our experiments demonstrate the effectiveness of facial beauty prediction tasks.
Keywords: Facial Beauty Prediction; Multi-Grained Scanning; Deep Forest; Cascade Forest;10,000 Facial Beauty Datasets.
Special Issue on: MISC'18 Recent Advancements in Intelligent Systems
Virtual Network Functions Placement in 5G Architecture: A Survey and a Multi-Agent Approach
by Sara Retal, Abdellah Idrissi
Abstract: New needs required by emerging communication obligations necessitate a fifth-generation (5G) mobile network. Indeed, the evolution of communication thanks to all connected devices made the society highly networked. Consequently, information and data must be available everywhere and every-time for everyone. 5G Mobile Network Architecture is a new framework that uses existing technologies such as Cloud computing to meet the requirements of numerous applications that increase data traffic daily. In this vein, we provide a comprehensive analysis of one of the main challenges facing this architecture, which is the placement of the virtual network functions (VNFs) and emphasize its link with the Cloud computing field. We also present significant works dealing with this problem, comparing them, and propose a multi-agent system approach that enhances these works.
Keywords: artificial intelligence; multi-agent system; virtual network functions placement; cloud computing.
New intelligent strategy for encryption decisional support system
by Salima TRICHNI, Fouzia OMARY, Abdellah IDRISSI, Mohammed BOUGRINE, Manar ABOUREZQ
Abstract: Despite the exponential evolution of information technologies and IT infrastructure, IT security tools (including encryption) do not always guarantee the same level of security. One can find very powerful encryption systems for images but not necessarily for videos or literal texts.
Our contribution is to achieve an intelligent encryption strategy for communication software, allowing the use of the most appropriate encryption system for the messages specificities, the infrastructure, and the destination platform. For this, we start with a classification of data to secure. Then, we establish a multidimensional data analysis through a Datawarehousse design in the star model. Moreover, in order to make the system intelligent, we populate the knowledge base with all the encryption results performed after each communication. Finally, we propose to apply the Block Nested Loop algorithm on ciphering algorithms skyline candidate to decide the one who meets most of the criteria to secure communication.
Keywords: IT security; Cryptography; Encryption; Communication Software; Artificial Intelligence; Multidimensional Data Analysis; Block Nested Loop; Skyline.
AI to prevent Cyber-violence: Harmful Behavior Detection in Social Media
by Randa Zarnoufi, Mehdi Boutbi, Mounia Abik
Abstract: Cyber-violence has recently emerged with an increased number of victims because of the expansion of technology use. It has a negative effect on victims who may suffer from significant physical and psychological pain, which has led to a significant amount of research in psychology and e-health to detect the act of cyber-violence. In the computational field, the majority of works are interested in multiple related aspects to cyber-violence, but no one of them, in our knowledge, has studied the perpetrator's harmful behavior from an emotional dimension. In this work, we focus on cyber perpetrator's emotional state. Our objective is to discover the relationship between the emotional state of social media users and the harmful behavior within the cyber-violence act. Our approach is based on supervised machine learning namely Ensemble Machine Learning and engineered features related to Plutchik wheel of basic emotions extracted with semantic similarity using word embedding. The findings show a significant association between the individuals emotional state and harmful intention. This result can be a good indicator of online users susceptibility to cyber-violence and therefore can help in dealing with it.
Keywords: Cyber-violence; E-health; Social Networks; Harmful Behavior; Emotional States; Engineered Features; Ensemble Machine Learning.
Semantic Approach Using Unified and Summurized Ontologies for Analyzing Data from Social Media
by Asmae EL KASSIRI, Fatima-zahra BELOUADHA
Abstract: Aggregating and analyzing Web social data is an important and interesting issue having an added value in various domains. Nevertheless, a major challenge to this issue is how to aggregate huge data scattered over a multitude of social media and be able to meet different analysis requirements and objectives such as recommendation, community detection, link prediction and sentiment analysis. In this context, we propose to use a summarized ontology of different inferred metrics that could be mutually reused to perform various analysis processes without redundant computing. According to the continuous evolution of Online Social Networks (OSN), these metrics are dynamically inferred from a unified semantic model that extends standard ontologies used in Social Web field. The proposed extension allows representing and aggregating data from a multitude of the most popular OSN.
Keywords: Online Social Networks; Semantic analysis; Ontologies; Data aggregation; Summarized data; inference and mapping rules.
An Efficient Privacy Solution for Electronic Health Records in Cloud Computing
by Mbarek Marwan, Fatima Sifou, Feda AlShahwan
Abstract: Healthcare organizations have shown a great interest in deploying electronic medical records (EMR) to save clinical data accurately and to represent the state of a patient over time. In this context, these tools are used to store digital data that include medical history, medication and allergies, laboratory test results and radiology reports. A crucial advantage of deploying an EMR system is that clients could more easily access and share clinical data. However, interoperability and the costs to maintain and operate these applications are considered the major obstacle to effectively using this concept. Using a cloud environment to build EMR systems is almost universally seen as the easiest and most practical option for the healthcare domain due to lower costs and scalability of cloud applications. Another advantage of cloud computing is outsourcing computation tasks to remote data centers that are managed by a third-party. Even with all of the above advantages of cloud computing, most existing solutions are not mature enough to provide adequate security mechanisms to protect the privacy of patients' health information. In fact, it is difficult to adequately implement traditional privacy mechanisms because of the trade-off between usability and confidentiality. In this respect, we propose a practical and highly secure solution to boost cloud services, especially image processing. Concretely, our approach uses a segmentation method to break up the secret data into small portions. Likewise, the proposal distributes the computations across many nodes; thereby optimizing the overall execution time and ensuring an adequate security level. Additionally, we propose a novel access control based on users trust and UMA framework. More interestingly, the trust-aware access control does not require the heavy computational load needed for complex operations as in traditional models. The implementation results prove the accuracy of this approach and show that this solution can help cloud providers to meet the stringent security standards and performance requirements.
Keywords: cloud computing; image processing; health records; security; K-means; access control.