Table of contents

PREFACE

AIAAT 2017 was held in Hawaii, USA during Aug.30-Sept.2, 2017. AIAAT 2017 was organized by Hong Kong Society of Mechanical Engineers (HKSME). The conference provides a useful and wide platform both for display the latest research and for exchange of research results and thoughts in Artificial Intelligence Applications and Technologies. The participants of the conference were from almost every part of the world, with background of either academia or industry, even well-known enterprise. The success and prosperity of the conference is reflected high level of the papers received.

The proceedings are a compilation of the accepted papers and represent an interesting outcome of the conference. This book covers 3 chapters: Machine Learning; Signal Processing and Control; Pattern Recognition.

We would like to acknowledge all of those who supported AIAAT 2017. Each individual and institutional help were very important for the success of this conference. Especially we would like to thank the organizing committee for their valuable advices in the organization and helpful peer review of the papers.

We sincerely hope that AIAAT 2017 will be a forum for excellent discussions that will put forward new ideas and promote collaborative researches. We are sure that the proceedings will serve as an important research source of references and the knowledge, which will lead to not only scientific and engineering progress but also other new products and processes.

Prof. Dan Zhang, York University, Canada

Prof. Du Ruxu, The Chinese University of Hong Kong, HK

List of Conference Chairman, Conference Committee Chair, Program Committee Chair, Publication Chair, Technical Program Committee are available in this PDF.

All papers published in this volume of IOP Conference Series: Materials Science and Engineering have been peer reviewed through processes administered by the proceedings Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

In this paper, a motion segmentation algorithm design is presented with the goal of segmenting a learned trajectory from demonstration such that each segment is locally maximally different from its neighbors. This segmentation is then exploited to appropriately scale (dilate/squeeze and/or rotate) a nominal trajectory learned from a few demonstrations on a fixed experimental setup such that it is applicable to different experimental settings without expanding the dataset and/or retraining the robot. The algorithm is computationally efficient in the sense that it allows facile transition between different environments. Experimental results using the Baxter robotic platform showcase the ability of the algorithm to accurately transfer a feeding task.

Extreme learning machine (ELM) is a widely used neural network with random weights (NNRW), which has made great contributions to many fields. However, the relationship between the parameters and the performance of ELM has not been fully investigated yet, i.e. the impact of the number of hidden layer nodes, the randomization range of the weights between the input layer and hidden layer, the randomization range of the threshold of hidden nodes, and the type of activation functions. In this paper, eight benchmark functions are used to study this relationship. We have some interesting findings, such as more hidden layer nodes cannot guarantee the best performance of ELM, the empirical randomization range of the hidden weights (i.e., [-1, 1]) and the empirical randomization range of the threshold of hidden nodes (i.e., [0, 1]) may not lead to the optimal performance of ELM models, and ELM with sigmoid as the activation function always achieves better performance on some regression problems than ELM with tribas as the activation function. We hope the findings from our work could provide a useful guidance for researchers to select right parameters for ELM.

Cyber-physical systems represent a new emerging field in automatic control. The fault system is a key component, because modern, large scale processes must meet high standards of performance, reliability and safety. Fault propagation in large scale chemical processes can lead to loss of production, energy, raw materials and even environmental hazard. The present paper develops a multi-agent fault-tolerant control architecture using robust fractional order controllers for a (¹³C) cryogenic separation column cascade. The JADE (Java Agent DEvelopment Framework) platform was used to implement the multi-agent fault tolerant control system while the operational model of the process was implemented in Matlab/SIMULINK environment. MACSimJX (Multiagent Control Using Simulink with Jade Extension) toolbox was used to link the control system and the process model. In order to verify the performance and to prove the feasibility of the proposed control architecture several fault simulation scenarios were performed.

Fixed point learning has become an important tool to analyse large scale distributed system such as urban traffic network. This paper presents a fixed point learning based intelligence traffic network control system. The system applies convergence property of fixed point theorem to optimize the traffic flow density. The intelligence traffic control system achieves maximum road resources usage by averaging traffic flow density among the traffic network. The intelligence traffic network control system is built based on decentralized structure and intelligence cooperation. No central control is needed to manage the system. The proposed system is simple, effective and feasible for practical use. The performance of the system is tested via theoretical proof and simulations. The results demonstrate that the system can effectively solve the traffic congestion problem and increase the vehicles average speed. It also proves that the system is flexible, reliable and feasible for practical use.

Industry 4.0 is one of the most formative terms in current times. Subject of research are particularly smart and autonomous mobile platforms, which enormously lighten the workload and optimize production processes. In order to interact with humans, the platforms need an in-depth knowledge of the environment. Hence, it is required to detect a variety of static and non-static objects. Goal of this paper is to propose an accurate and real-time capable object detection and localization approach for the use on mobile platforms. A method is introduced to use the powerful detection capabilities of a neural network for the localization of objects. Therefore, detection information of a neural network is combined with depth information from a RGB-D camera, which is mounted on a mobile platform. As detection network, YOLO Version 2 (YOLOv2) is used on a mobile robot. In order to find the detected object in the depth image, the bounding boxes, predicted by YOLOv2, are mapped to the corresponding regions in the depth image. This provides a powerful and extremely fast approach for establishing a real-time-capable Object Locator. In the evaluation part, the localization approach turns out to be very accurate. Nevertheless, it is dependent on the detected object itself and some additional parameters, which are analysed in this paper.

This paper investigates the application of a novel method for classification called Feature Weighted Self Organizing Map (FWSOM) that analyses the topology information of a converged standard Self Organizing Map (SOM) to automatically guide the selection of important inputs during training for improved classification of data with redundant inputs, examined against two traditional approaches namely neural networks and Support Vector Machines (SVM) for the classification of EEG data as presented in previous work. In particular, the novel method looks to identify the features that are important for classification automatically, and in this way the important features can be used to improve the diagnostic ability of any of the above methods. The paper presents the results and shows how the automated identification of the important features successfully identified the important features in the dataset and how this results in an improvement of the classification results for all methods apart from linear discriminatory methods which cannot separate the underlying nonlinear relationship in the data. The FWSOM in addition to achieving higher classification accuracy has given insights into what features are important in the classification of each class (left and right-hand movements), and these are corroborated by already published work in this area.

This paper investigates a mathematical model inspired by nature, and presents a Meta-Heuristic that is efficient in improving the performance of an informed search, when using strategy A * using a General Search Tree as data structure. The work hypothesis suggests that the investigated meta-heuristic is optimal in nature and may be promising in minimizing the computational resources required by an objective-based agent in solving high computational complexity problems (n-part puzzle) as well as In the optimization of objective functions for local search agents. The objective of this work is to describe qualitatively the characteristics and properties of the mathematical model investigated, correlating the main concepts of the A * function with the significant variables of the metaheuristic used. The article shows that the amount of memory required to perform this search when using the metaheuristic is less than using the A * function to evaluate the nodes of a general search tree for the eight-piece puzzle. It is concluded that the meta-heuristic must be parameterized according to the chosen heuristic and the level of the tree that contains the possible solutions to the chosen problem.

Sentence classification is one of the significant issues in Natural Language Processing (NLP). Feature extraction is often regarded as the key point for natural language processing. Traditional ways based on machine learning can not take high level features into consideration, such as Naive Bayesian Model. The neural network for sentence classification can make use of contextual information to achieve greater results in sentence classification tasks. In this paper, we focus on classifying Chinese sentences. And the most important is that we post a novel architecture of Convolutional Neural Network (CNN) to apply on Chinese sentence classification. In particular, most of the previous methods often use softmax classifier for prediction, we embed a linear support vector machine to substitute softmax in the deep neural network model, minimizing a margin-based loss to get a better result. And we use tanh as an activation function, instead of ReLU. The CNN model improve the result of Chinese sentence classification tasks. Experimental results on the Chinese news title database validate the effectiveness of our model.

Machine Learning applied to Automatic Audio Surveillance has been attracting increasing attention in recent years. In spite of several investigations based on a large number of different approaches, little attention had been paid to the environmental temporal evolution of the input signal. In this work, we propose an exploration in this direction comparing the temporal correlations extracted at the feature level with the one learned by a representational structure. To this aim we analysed the prediction performances of a Recurrent Neural Network architecture varying the length of the processed input sequence and the size of the time window used in the feature extraction. Results corroborated the hypothesis that sequential models work better when dealing with data characterized by temporal order. However, so far the optimization of the temporal dimension remains an open issue.

This paper proposes an approach to smooth optimal control design of two-point boundary value problem of switched systems with fuzzy operating regions. The switched system is modeled via a T-S fuzzy system, and then the fuzzy inference method of T-S fuzzy system is used to transform the switched system to a nonlinear system. For this nonlinear system, an optimal controller is designed. The control signal is a fuzzy control with weighted average defuzzifier. The fuzzy sets are used to partition the time space and a Particle Swarm Optimization (PSO) algorithm is proposed to optimize the weights. Simulations results demonstrate the applicability and effectiveness of the proposed method.

Spectrum sensing (SS) has attracted much concern of researchers due to its significant contribution on the spectral efficiency. Energy Detection (ED) has been a critical method for Spectrum Sensing in Cognitive Radio Networks (CRNS) due to its low complexity and simple implement. However, noise uncertainty in ED greatly degrades the detection performance, especially under a low Signal-to-Noise Ratio (SNR). To remove noise uncertainty as much as possible, a scheme based on subspace decomposition is proposed for SS, where the received signal is decomposed into two parts: noise subspace and signal-plus-noise subspace. Then the closed-form solution of the detection and false alarm probabilities is given on the basis of the signal-plus-noise subspace in Rayleigh fading channel. The energy of the remainders after removal of noise subspace and noise contribution in signal-plus-noise subspace is used to decide whether the primary user (PU) exists by a comparison with a redesigned threshold. Eventually, some simulations based on MATLAB platform is made to validate the proposed method.

One of the critical topics in medical low-dose Computed Tomography (CT) imaging is how best to maintain image quality. As the quality of images decreases with lowering the X-ray radiation dose, improving image quality is extremely important and challenging. We have proposed a novel approach to denoise low-dose CT images. Our algorithm learns directly from an end-to-end mapping from the low-dose Computed Tomography images for denoising the normal-dose CT images. Our method is based on a deep convolutional neural network with rectified linear units. By learning various low-level to high-level features from a low-dose image the proposed algorithm is capable of creating a high-quality denoised image. We demonstrate the superiority of our technique by comparing the results with two other state-of-the-art methods in terms of the peak signal to noise ratio, root mean square error, and a structural similarity index.

Obstacle detection and avoidance are major research fields in unmanned aviation. Map based obstacle detection approaches often use discrete world representations such as probabilistic grid maps to fuse incremental environment data from different views or sensors to build a comprehensive representation. The integration of continuous measurements into a discrete representation can result in rounding errors which, in turn, leads to differences between the artificial model and real environment. The cause of these deviations is a low spatial resolution of the world representation comparison to the used sensor data. Differences between artificial representations which are used for path planning or obstacle avoidance and the real world can lead to unexpected behavior up to collisions with unmapped obstacles. This paper presents three approaches to the treatment of errors that can occur during the integration of continuous laser measurement in the discrete probabilistic grid. Further, the quality of the error prevention and the processing performance are compared with real sensor data.

3D reconstruction is of vital importance to detect and monitor the underwater environment. A method based on geometric transformation of mechanical scanning sonar and depth information is proposed, in which the point cloud data from sonar and depth gauge are acquired to reconstruct the underwater 3D environment. However, noise and interference can affect the measurement of sonar, and movement of sonar during measurement can lead to distortion of the received data. Meanwhile, translation and rotation movement of sonar head may happen when ROV dives which can lead to different body reference coordinates of different scanning. To solve this, pre-processing and motion compensation are implemented at first, and underwater matching correction algorithm is used to calculate the translation and rotation of the sonar head. Then the inverse operation is implemented to convert the scan data of every depth into the same coordinate reference system. Finally, surface reconstruction of point clouds from sonar the depth information are used to reconstruct underwater environment based on MLS (Moving Least Square Method) using PCL (Point Cloud Library). Water tank experiments verify the effectiveness of the proposed method.

Tokamak is a fusion reactor that provides extremely high temperature needed for fusion process. Such a temperature is created by confining plasma via some poloidal coils. Control of voltages to be applied to the poloidal coils is an important control problem for proper operation of the overall system. This paper deals with the nonlinear robust control of plasma current, shape, and position in a Tokamak reactor. The paper briefly discusses the modeling issue of a Tokamak, and then, based on a nonlinear state-space model, presents three different versions of a robust controller for the process. All parameters of the system are assumed to be unknown for all three types of the controllers. Lyapunov-type arguments are used to prove the stability of the proposed control schemes. Advantages and disadvantages with each type of controllers are discussed in detail.

In this paper we present and evaluate a cognitive computing approach for classification of dissatisfaction and four complaint specific complaint classes in correspondence documents between insurance clients and an insurance company. A cognitive computing approach includes the combination classical natural language processing methods, machine learning algorithms and the evaluation of hypothesis. The approach combines a MaxEnt machine learning algorithm with language modelling, tf-idf and sentiment analytics to create a multi-label text classification model. The result is trained and tested with a set of 2500 original insurance communication documents written in German, which have been manually annotated by the partnering insurance company. With a F1-Score of 0.9, a reliable text classification component has been implemented and evaluated. A final outlook towards a cognitive computing insurance assistant is given in the end.

With the recent increase in data online, discovering meaningful opportunities can be time-consuming and complicated for many individuals. To overcome this data overload challenge, we present a novel text-content-based recommender system as a valuable tool to predict user interests. To that end, we develop a specific procedure to create user models and item feature-vectors, where items are described in free text. The user model is generated by soliciting from a user a few keywords and expanding those keywords into a list of weighted near-synonyms. The item feature-vectors are generated from the textual descriptions of the items, using modified tf-idf values of the users' keywords and their near-synonyms. Once the users are modeled and the items are abstracted into feature vectors, the system returns the maximum-similarity items as recommendations to that user. Our experimental evaluation shows that our method of creating the user models and item feature-vectors resulted in higher precision and accuracy in comparison to well-known feature-vector-generating methods like Glove and Word2Vec. It also shows that stemming and the use of a modified version of tf-idf increase the accuracy and precision by 2% and 3%, respectively, compared to non-stemming and the standard tf-idf definition. Moreover, the evaluation results show that updating the user model from usage histories improves the precision and accuracy of the system. This recommender system has been developed as part of the Agnes application, which runs on iOS and Android platforms and is accessible through the Agnes website.

Short text is considerably different from traditional long text documents due to its shortness and conciseness, which somehow hinders the applications of conventional machine learning and data mining algorithms in short text classification. According to traditional artificial intelligence methods, we divide short text classification into three steps, namely preprocessing, feature selection and classifier comparison. In this paper, we have illustrated step-by-step how we approach our goals. Specifically, in feature selection, we compared the performance and robustness of the four methods of one-hot encoding, tf-idf weighting, word2vec and paragraph2vec, and in the classification part, we deliberately chose and compared Naive Bayes, Logistic Regression, Support Vector Machine, K-nearest Neighbor and Decision Tree as our classifiers. Then, we compared and analysed the classifiers horizontally with each other and vertically with feature selections. Regarding the datasets, we crawled more than 400,000 short text files from Shanghai and Shenzhen Stock Exchanges and manually labeled them into two classes, the big and the small. There are eight labels in the big class, and 59 labels in the small class.

Most recommendation systems in the major electronic commerce platforms are influenced by the long tail effect more or less. There are sufficient researches of how to assess recommendation effect while no criteria to evaluate long tail recommendation rate. In this study, we first discussed the existing problems of recommending long tail products through specific experiments. Then we proposed a long tail evaluation criteria and compared the performance in long tail recommendation between different models.

The technology of computer vision is used in the training of military shooting. In order to overcome the limitation of the bullet holes recognition using Video Image Analysis that exists over-detection or leak-detection, this paper adopts the support vector machine algorithm and convolutional neural network to extract and recognize Bullet Holes in the digital video and compares their performance. It extracts HOG characteristics of bullet holes and train SVM classifier quickly, though the target is under outdoor environment. Experiments show that support vector machine algorithm used in this paper realize a fast and efficient extraction and recognition of bullet holes, improving the efficiency of shooting training.

In order to achieve a more convenient and accurate digital museum navigation, we have developed a real–time and online-to-offline museum exhibits recognition system using image recognition method based on deep learning. In this paper, the client and server of the system are separated and connected through the HTTP. Firstly, by using the client app in the Android mobile phone, the user can take pictures and upload them to the server. Secondly, the features of the picture are extracted using the deep learning network in the server. With the help of the features, the pictures user uploaded are classified with a well-trained SVM. Finally, the classification results are sent to the client and the detailed exhibition's introduction corresponding to the classification results are shown in the client app. Experimental results demonstrate that the recognition accuracy is close to 100% and the computing time from the image uploading to the exhibit information show is less than 1S. By means of exhibition image recognition algorithm, our implemented exhibits recognition system can combine online detailed exhibition information to the user in the offline exhibition hall so as to achieve better digital navigation.

Mapping customer requirements into product configurations is a crucial step for product design, while, customers express their needs ambiguously and locally due to the lack of domain knowledge. Thus the data mining process of customer requirements might result in fragmental information with high dimensional sparsity, leading the mapping procedure risk uncertainty and complexity. The Expert Judgment is widely applied against that background since there is no formal requirements for systematic or structural data. However, there are concerns on the repeatability and bias for Expert Judgment. In this study, an integrated method by adjusted Local Linear Embedding (LLE) and Naïve Bayes (NB) classifier is proposed to map high dimensional sparse customer requirements to product configurations. The integrated method adjusts classical LLE to preprocess high dimensional sparse dataset to satisfy the prerequisite of NB for classifying different customer requirements to corresponding product configurations. Compared with Expert Judgment, the adjusted LLE with NB performs much better in a real-world Tablet PC design case both in accuracy and robustness.

The sentiments analysis of product reviews is designed to help customers understand the status of the product. The traditional method of sentiments analysis relies on the input of a fixed feature vector which is performance bottleneck of the basic codec architecture. In this paper, we propose an attention mechanism with BRNN-CNN model, referring to as ARCNN model. In order to have a good analysis of the semantic relations between words and solves the problem of dimension disaster, we use the GloVe algorithm to train the vector representations for words. Then, ARCNN model is proposed to deal with the problem of deep features training. Specifically, BRNN model is proposed to investigate non-fixed-length vectors and keep time series information perfectly and CNN can study more connection of deep semantic links. Moreover, the attention mechanism can automatically learn from the data and optimize the allocation of weights. Finally, a softmax classifier is designed to complete the sentiment classification of reviews. Experiments show that the proposed method can improve the accuracy of sentiment classification compared with benchmark methods.