The wood-based furniture manufacturing industries prioritize quality of production to meet higher market demands. Identifying various types of edge-glued wooden panel defects are a challenge for a human worker or a camera. Several studies have shown that the detection of edge-glued defects with low, high, normal, overlong, short is identified but detection of residue and bluntness is highly challenging. Thus, the present model identifies defects of low, high, normal, overlong, short by computer vision and/or deep learning, whereas defects of residue and bluntness by deep learning based decide by pass for having better performance. The goal of this paper is to provide an improved defect detection solution for wood-based furniture manufacturing industries by process automation. Therefore, a system was designed that takes defect input images from a camera as raw image and laser-aligned image for defect detection of the edge-glued wooden panel. The process automation then performs computer vision-based image features extraction with deep learning for defect detection. The aim of this paper is to solve edge-glued defect detection problems by using design and implementation of edge-glued wooden defect detection, that can be stated as edge-glued wooden panel defect detection using deep learning (WDD-DL) for process automation by artificial intelligence and Automated Optical Inspection (AOI) consolidation. Possibly there exist several types of defects on the edges while edge-banding on the wooden panel in furniture manufacturing. Therefore, the scope is to achieve higher accuracy by raw image and laser-aligned image feature extraction using deep learning algorithms for final result defect classification in WDD-DL by AOI. The WDD-DL system uses Gabor, Harris corner, morphology, structured light detection and curvature calculation for pre-processing and InceptionResnetV2 Convolutional Neural Network algorithm to attain the best results. The applications of this work can be found in quality control of the furniture manufacturing industry for an edge, corner, joint defect detection of the wooden panels. The WDD-DL achieves best results as the precision, recall and F1 score are 0.97, 0.90 and 0.92, respectively. The experiments demonstrate higher accuracy achievement as compared to other methods with overkill and escape rate analysis. Ultimately, the discussion section provides an interesting experience sharing about the necessary factors for implementing the WDD-DL in real-time industrial operations.

BACKGROUND:Antinuclear antibody pattern recognition is vital for autoimmune disease diagnosis but labor-intensive for manual interpretation. To develop an automated pattern recognition system, we established machine learning models based on the International Consensus on Antinuclear Antibody Patterns (ICAP) at a competent level, mixed patterns recognition, and evaluated their consistency with human reading.METHODS:51,694 human epithelial cells (HEp-2) cell images with patterns assigned by experienced medical technologists collected in a medical center were used to train six machine learning algorithms and were compared by their performance. Next, we choose the best performing model to test the consistency with five experienced readers and two beginners.RESULTS:The mean F1 score in each classification of the best performing model was 0.86 evaluated by Testing Data 1. For the inter-observer agreement test on Testing Data 2, the average agreement was 0.849 (κ) among five experienced readers, 0.844 between the best performing model and experienced readers, 0.528 between experienced readers and beginners. The results indicate that the proposed model outperformed beginners and achieved an excellent agreement with experienced readers.CONCLUSIONS:This study demonstrated that the developed model could reach an excellent agreement with experienced human readers using machine learning methods.

Early prediction of clinical deterioration such as adverse events (AEs), improves patient safety. National Early Warning Score (NEWS) is widely used to predict AEs based on the aggregation of 6 physiological parameters. We took the same parameters as the features for AE prediction using deep learning algorithms (AEP-DLA) among hospitalized adult patients. The aim of this study is to get better performance than traditional naïve mathematical calculations by introducing novel vital sign data preprocessing schemes. We retrospectively collected the data from our electronic medical record data warehouse (2007 ~ 2017). AE rate of all 99,861 admissions was 6.2%. The dataset was divided into training and testing datasets from 2007–2015 and 2016–2017 respectively. In real-life clinical care, physiological parameters were not recorded every hour and missed frequently, for example, Glasgow Coma Scale (GCS). The expert domain suggested that missed GCS was rated as 15. We took two strategies (stack series records and align by hour) in the data preprocessing and tripling the values of negative samples for class balancing (CB). We used the last 28 hours' serial data to predict AEs 3 hours later with Random Forest, XGBoost, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). It is shown that CNN with CB and align by hour got the best results comparing to the other methods. The precision, recall and area under curve were 0.841, 0.928 and 0.995 respectively. The performance of the model is also better than those proposed in the published literatures.

Performance problems in a cloud service are difficult to diagnose because they may be caused by various system components. This study proposes an automatic cloud service testing and bottleneck detection system that is applicable to different types of services. With the proposed test module, the user can customize test scenarios to automatically test and collect the corresponding metrics of the target service. Afterward, the proposed bottleneck detection algorithm analyzes the collected metrics and determines whether a bottleneck is presented in the target system. The bottleneck detection module also provides a scaling recommendation for the service provider to facilitate the service system reconfiguration. The experimental results reveal that the proposed system could detect a potential bottleneck in a service system accurately. In accordance with the scaling recommendation, the performance of the target cloud service can be improved efficiently after reconfiguration. Therefore, usage of the proposed system can ensure a high quality of service, and the service level objective could be fulfilled.

Cloud testing has become a crucial concern due to the explosion of cloud services and requirements of reliable and high-quality service. The technologies of virtualization and infrastructures of cloud services have enabled service providers to deploy their services conveniently. However, current test models decrease the flexibility of test scenarios and test client customization. For solving these problems, this study proposes a testing service in the cloud to help the testing environment generation, deployment and execution of test scenarios. The tester can customize test scenarios involving clients, network topologies, and test scripts. In the paper, a demonstration test toolkit was implemented, and also used for testing distributed file system with various numbers of hosts, threads, and virtual network topologies.

With the rapid growth of mobile applications or web services, many users are using the same service at the same time which leads to the need for scalable application services. To enlarge the capability of handling large volumes of concurrent requests, server-side push technologies have been then introduced recently. Server-side push technology tries to eliminate unnecessary client requests by sending notification messages to clients when a data change event is triggered on the server side. There are two famous server-side push frameworks used today, and they are services using the Pub/Sub protocol, and the Ajax Push Model. These two mechanisms need middleware between clients and servers which result in complex and heavy-weight system architectures for application developers. To simplify development and increase ability to accommodate the growth in application users, a RESTful notification service is proposed in this paper. The notification service leverages the event-driven characteristic of JavaScript, and pushes response data asynchronously to different requests. Based on the RESTful software architecture style, the proposed notification service is not only a lightweight system but also has impressive performance.

Cloud service testing ensures that services run properly and meet the Service Level Agreement (SLA) requirements. However, performance problems of a cloud service, such as the availability and reliability, are difficult to diagnose because these issues might be caused from different system components. To solve these problems, this study proposes an architecture for testing environment configuration and quality estimation for both of fault diagnosis and bottleneck detection. The proposed system has two components: offline testing and online management. In offline testing, target service are tested by using the proposed testing module and corresponding metrics are collected for further analysis. The analyzed results which are separated between fault diagnosis and bottleneck detection will be stored in knowledge databases. Finally, online management module will automatically suggest how to do when facing this kind of problem according to knowledge based diagnosis.

Decision tree is one of the famous classification methods in data mining. Many researches have been proposed, which were focusing on improving the performance of decision tree. However, those algorithms are developed and run on traditional distributed systems. Obviously the latency could not be improved while processing huge data generated by ubiquitous sensing node in the era without new technology help. In order to improve data processing latency in huge data mining, in this paper, we design and implement a new parallelized decision tree algorithm on a CUDA (compute unified device architecture), which is a GPGPU solution provided by NVIDIA. In the proposed system, CPU is responsible for flow control while the GPU is responsible for computation. We have conducted many experiments to evaluate system performance of CUDT and made a comparison with traditional CPU version. The results show that CUDT is 5 ∼ 55 times faster than Weka-j48 and is 18 times speedup than SPRINT for large data set.

Recommendation systems are majorly developed based on relationships of product features or between consumer attributes. Most of them need a lot of analysis of historical shopping transactions and statistical user or product features to come out good suggestions for consumers to make right decisions. However, it does not fit into the users' shopping experiences for specialty stores of sporting goods. The characteristics of sporting goods specialty stores are less products and less volume of customers than other types of stores. It is hard for recommender systems to help users making the shopping decisions with limited product information and users' historical shopping behaviors. It is the purpose of this paper to propose a two-phase recommendation technique based on the AHP methodology to improve the selling of sporting goods specialty stores. We also implemented a practice system for a specialty store selling badminton-related goods. The results show that it is easier for sporting goods stores to promote products, and help consumers to choose products based on their own features.

Recently, depression has becomes a widespread disease throughout the world. However, most people are not aware of the possibility of becoming depressed during their daily lives. Therefore, obtaining an accurate diagnosis of depression is an important issue in healthcare. In this study, we built an inference model based on an ontology and a Bayesian network to infer the possibility of becoming depressed, and we implemented a prototype using a mobile agent platform as a proof-of-concept in the mobile cloud. We developed an ontology model based on the terminology used to describe depression and we utilized a Bayesian network to infer the probability of becoming depressed. We also implemented the system using multi-agents to run on the Android platform, thereby demonstrating the feasibility of this method, and we addressed various implementation issues. The results showed that our method may be useful for inferring a diagnosis of depression.

File systems are designed to control how files are stored and retrieved. Without knowing the context and semantics of file contents, file systems often contain duplicate copies and result in redundant consumptions of storage space and network bandwidth. It has been a complex and challenging issue for enterprises to seek deduplication technologies to reduce cost and increase the storage efficiency. To solve such problem, researchers proposed in-line or offline solutions for primary storages or backup systems at the subfile or whole-file level. Some of the technologies are used for file servers and database systems. Fewer studies focus on the cloud file system deduplication technologies at the application level, especially for the Hadoop distributed file system. It is the goal of this paper to design a file deduplication framework on Hadoop distributed file system for cloud application developers. The architecture, interface, and implementation experiences are also shared in this paper.

Video surveillance system (VSS) is increasingly becoming important part in daily life. For traditional VSS, each camera stores streaming data to a centralized server. It will create a great volume of video data for a large VSS and may raise some issues in keeping daily video data to centralized server, such as limited bandwidth and storage insufficiency of server and lower reliability and scalability. In order to solve the problem, we have proposed an architecture for VSS based on well-developed peer-to-peer technique and emerging cloud computing. In this paper, we implement a large-scale VSS (LVSS) and propose an evaluation framework based on the proposed architecture. We implement the LVSS exploits inherent characteristics of P2P and cloud computing to provide an economic, scalable, reliable, and efficient approach to store streaming video. We conducted some experiments to evaluate effectiveness and efficiency of the system. The result shows that the proposed architecture outperforms traditional VSS in terms of effectiveness, efficiency, reliability, and scalability.

A hybrid cloud integrates private clouds and public clouds into one unified environment. For the economy and the efficiency reasons, the hybrid cloud environment should be able to automatically maximize the utilization rate of the private cloud and minimize the cost of the public cloud when users submit their computing jobs to the environment. In this paper, we propose the Adaptive-Scheduling-with-QoS-Satisfaction algorithm, namely AsQ, for the hybrid cloud environment to raise the resource utilization rate of the private cloud and to diminish task response time as much as possible. We exploit runtime estimation and several fast scheduling strategies for near-optimal resource allocation, which results in high resource utilization rate and low execution time in the private cloud. Moreover, the near-optimal allocation in the private cloud can reduce the amount of tasks that need to be executed on the public cloud to satisfy their deadline. For the tasks that have to be dispatched to the public cloud, we choose the minimal cost strategy to reduce the cost of using public clouds based on the characteristics of tasks such as workload size and data size. Therefore, the AsQ can achieve a total optimization regarding cost and deadline constraints. Many experiments have been conducted to evaluate the performance of the proposed AsQ. The results show that the performance of the proposed AsQ is superior to recent similar algorithms in terms of task waiting time, task execution time and task finish time. The results also show that the proposed algorithm achieves a better QoS satisfaction rate than other similar studies.

Recommendation system is an important method of solving the problem of information overload. It also helps consumers to save time while searching for goods. Numerous recommendation techniques are proposed. However, they still have to confront some weaknesses such as cold-start, gray sheep and matrix sparsity problems. The purpose of this paper is to propose a method to overcome the cold-start problem and recommend a fit item for consumers to improve the personalized service. The proposed method can be applied in the e-commerce websites of exclusive or specialty stores. It is a combination of the product knowledge and Analytic Hierarchy Process (AHP) method. There are two phases in the proposed method. Phase 1 is to calculate the weight between product attributes and create a candidate product set. Phase 2 is to conduct the recommendation from the candidate set. This paper also introduces the implementation experiences by taking the badminton racket recommendation as a case study example.

Web Services are emerging technologies that enable application to application communication and reuse of autonomous services over Web. Composition of web services is a concept of integrating individual web services to conduct complex business transactions based on functionality and performance constraints. To simplify composite web service development, technologies of quality of service (QoS) based web services composition (QWSC) are widely used to build complex applications by discovering the best-fit web services. In this paper, a web service composition system is proposed based on our previous study, a Relaxable QoS-based service selection algorithm (RQSS), to demonstrate its feasibility and performance as well as to share the experiences of developing a web service composition system. The proposed system is compatible with WS-BPEL standards, which can be used for composite web application development through intuitive graphic user interfaces. The experimental results show that the proposed relaxable web composition system performs well with notably improvement for application availability and reliability.

Previous application sharing systems have no unified communication protocols. The well known application sharing protocols include RFB (Remote FrameBuffer), RDP (Remote Desktop Protocol) and ITU-T T.128. A generic application sharing architecture is proposed for providing the capability of connecting, various application sharing servers with the same application sharing client. A communication standard for application sharing systems is defined based on the Message-Oriented Middleware (MOM). To compare with previous application sharing systems, there are many advantages including easy to use, easy to connect, easy to integrate, higher scalability and less bandwidth consumption.

Locations of moving or missing objects are getting important information for context-aware applications which try to get the locality of an object, and then provide services pertaining to the object. To position an object, most systems use a predefined coordinate to compute object location while sensing the appearance of the target object. Usually, it is troublesome and costs much to define the base coordinate in advance for most object positioning systems, especially when the target object is locating in an unknown environment. To reduce the cost and complexity of object locating system and improve the accuracy of location, this paper proposed an Object Finding System based on RFID technology to identify the localities of target objects in buildings. In this paper, we introduce the design concepts of the proposed system as well as the algorithms used to calculate the object locations. In addition, the experimental results show that it is a feasibility study.

Efficient parallel algorithms for finding an optimum path-shaped or tree-shaped facility with a specified size in a tree network are presented. Four kinds of optimization criteria are considered: minimum eccentricity, maximum eccentricity, minimum distancesum, and maximum distancesum. Eight cases are obtained by combining facility shapes and optimization criteria. Optimal parallel algorithms for finding a minimum/maximum distancesum path were presented by B.C. Tansel et al. (1983). The other six cases are studied here. Two of them can be solved optimally in linear time*processor complexity. For the problem of finding a maximum distancesum tree, the algorithm presented is a polynomial time resolution.< >

There are two major parts to this dissertation. In the first part, efficient sequential and parallel algorithms for locating structured facilities on tree networks are developed. Both path-shaped and tree-shaped structure are addressed. For optimality criteria, minimum/maximum eccentricity/distancesum are considered. Among these problems, the case of finding the maximum distancesum tree of a specified size was unsolved. In 1985, Minieka stated that the maximum distancesum tree problem has computational complexity $O(2\sp{n})$, where n is the number of vertices in the tree network. We present a polynominal time sequential algorithm and a poly-log time parallel algorithm for the problem, thereby solving this open problem. For other problems, our algorithms improve the best known results significantly in terms of time-processor complexities. In the second part, we present an efficient scheme to allocate task graphs to Hypercube multiprocessors. Both precedence relations and communication requirements among tasks are considered in the allocation scheme. The optimality criterion is to minimize the turnaround time. Turnaround time includes processing time of tasks and communication time among tasks. The task graph is first translated into a MPE (Multistage PE) model, developed in this research, using the priority listing method to minimize the processing time of tasks. Then mappings between every stage of the MPE model and the underlying system, i.e. Hypercube, are performed. The mappings are performed by a heuristic branch and bound search algorithm with implicit depth first (LCBB/IDF). The experiments are conducted in a transputer net with 16 IMS T800 transputers. The results show that the allocation scheme is efficient and performs well with respect to search pruning and minimizing the turnaround time for different sizes and shapes of task graphs. ftn$\sp*$This work was partially supported by NATO under grant CRG890612 and NSF under grant CCR-9002351.