In recent years, deep learning based image recognition, speech recognition, text translation and other related applications have brought great convenience to people's lives. With the advent of the era of internet of everything, how to run a computationally intensive deep learning algorithm on a limited resources edge device is a major challenge. For an edge oriented computing vector processor, combined with a specific neural network model, a new data layout method for putting the input feature maps in DDR, rearrangement of the convolutional kernel parameters in the nuclear memory bank is proposed. Aiming at the difficulty of parallelism of two-dimensional matrix convolution, a method of parallelizing the matrix convolution calculation in the third dimension is proposed, by setting the vector register with zero as the initial value of the max pooling to fuse the rectified linear unit (ReLU) activation function and pooling operations to reduce the repeated access to intermediate data. On the basis of single core implementation, a multi-core implementation scheme of Inception structure is proposed. Finally, based on the proposed vectorization method, we realize five kinds of neural network models, namely, AlexNet, VGG16, VGG19, GoogLeNet, ResNet18, and performance statistics and analysis based on CPU, gtx1080TI and FT2000 are presented. Experimental results show that the vector processor has better computing advantages than CPU and GPU, and can calculate large-scale neural network model in real time.

We prove generalization error bounds of classes of low-rank matrices with some norm constraints for collaborative filtering tasks. Our bounds are tighter, compared to known bounds using rank or the related quantity only, by taking the additional L₁ and L_∞ constraints into account. Also, we show that our bounds on the Rademacher complexity of the classes are optimal.

During the execution of software, tremendous amounts of data can be recorded. By exploiting the execution data, one can discover behavioral models to describe the actual software execution. As a well-known open-source process mining toolkit, ProM integrates quantities of process mining techniques and enjoys a variety of applications in a broad range of areas. How to develop a better ProM software, both from user experience and software performance perspective, are of vital importance. To achieve this goal, we need to investigate the real execution behavior of ProM which can provide useful insights on its usage and how it responds to user operations. This paper aims to propose an effective approach to solve this problem. To this end, we first instrument existing ProM framework to capture execution logs without changing its architecture. Then a two-layered framework is introduced to support accurate ProM behavior discovery by characterizing both user interaction behavior and plug-in calling behavior separately. Next, detailed discovery techniques to obtain user interaction behavior model and plug-in calling behavior model are proposed. All proposed approaches have been implemented.

Many software development teams usually tend to focus on maintenance activities in general. Recently, many studies on bug severity prediction have been proposed to help a bug reporter determine severity. But they do not consider the reporter's expression of emotion appearing in the bug report when they predict the bug severity level. In this paper, we propose a novel approach to severity prediction for reported bugs by using emotion similarity. First, we do not only compute an emotion-word probability vector by using smoothed unigram model (UM), but we also use the new bug report to find similar-emotion bug reports with Kullback-Leibler divergence (KL-divergence). Then, we introduce a new algorithm, Emotion Similarity (ES)-Multinomial, which modifies the original Naïve Bayes Multinomial algorithm. We train the model with emotion bug reports by using ES-Multinomial. Finally, we can predict the bug severity level in the new bug report. To compare the performance in bug severity prediction, we select related studies including Emotion Words-based Dictionary (EWD)-Multinomial, Naïve Bayes Multinomial, and another study as baseline approaches in open source projects (e.g., Eclipse, GNU, JBoss, Mozilla, and WireShark). The results show that our approach outperforms the baselines, and can reflect reporters' emotional expressions during the bug reporting.

Recommender systems (RS) exploit user ratings on items and side information to make personalized recommendations. In order to recommend the right products to users, RS must accurately model the implicit preferences of each user and the properties of each product. In reality, both user preferences and item properties are changing dynamically over time, so treating the historical decisions of a user or the received comments of an item as static is inappropriate. Besides, the review text accompanied with a rating score can help us to understand why a user likes or dislikes an item, so temporal dynamics and text information in reviews are important side information for recommender systems. Moreover, compared with the large number of available items, the number of items a user can buy is very limited, which is called the sparsity problem. In order to solve this problem, utilizing item correlation provides a promising solution. Although famous methods like TimeSVD++, TopicMF and CoFactor partially take temporal dynamics, reviews and correlation into consideration, none of them combine these information together for accurate recommendation. Therefore, in this paper we propose a novel combined model called TmRevCo which is based on matrix factorization. Our model combines the dynamic user factor of TimeSVD++ with the hidden topic of each review text mined by the topic model of TopicMF through a new transformation function. Meanwhile, to support our five-scoring datasets, we use a more appropriate item correlation measure in CoFactor and associate the item factors of CoFactor with that of matrix factorization. Our model comprehensively combines the temporal dynamics, review information and item correlation simultaneously. Experimental results on three real-world datasets show that our proposed model leads to significant improvement compared with the baseline methods.

With the expansion of the Internet, the number of available Web services has increased. Web service clustering to identify functionally similar clusters has become a major approach to the efficient discovery of suitable Web services. In this study, we propose a Web service clustering approach that uses novel ontology learning and a similarity calculation method based on the specificity of an ontology in a domain with respect to information theory. Instead of using traditional methods, we generate the ontology using a novel method that considers the specificity and similarity of terms. The specificity of a term describes the amount of domain-specific information contained in that term. Although general terms contain little domain-specific information, specific terms may contain much more domain-related information. The generated ontology is used in the similarity calculations. New logic-based filters are introduced for the similarity-calculation procedure. If similarity calculations using the specified filters fail, then information-retrieval-based methods are applied to the similarity calculations. Finally, an agglomerative clustering algorithm, based on the calculated similarity values, is used for the clustering. We achieved highly efficient and accurate results with this clustering approach, as measured by improved average precision, recall, F-measure, purity and entropy values. According to the results, specificity of terms plays a major role when classifying domain information. Our novel ontology-based clustering approach outperforms comparable existing approaches that do not consider the specificity of terms.

Software-defined networking (SDN) is recognized as the next-generation networking paradigm. The software-defined architecture for underwater acoustic sensor networks (SDUASNs) has become a hot topic. However, the current researches on SDUASNs is still in its infancy, which mainly focuses on network architecture, data transmission and routing. There exists some shortcomings that the scale of the SDUASNs is difficult to expand, and the security maintenance is seldom dabble. Therefore, a scalable software-definition architecture for underwater acoustic sensor networks (SSDUASNs) is introduced in this paper. It realizes an organic combination of the knowledge level, control level, and data level. The new nodes can easily access the network, which could be conducive to large-scale deployment. Then, the basic security authentication mechanism called BSAM is designed based on our architecture. In order to reflect the advantages of flexible and programmable in SSDUASNs, security authentication mechanism with pre-push (SAM-PP) is proposed in the further. In the current UASNs, nodes authentication protocol is inefficient as high consumption and long delay. In addition, it is difficult to adapt to the dynamic environment. The two mechanisms can effectively solve these problems. Compared to some existing schemes, BSAM and SAM-PP can effectively distinguish between legal nodes and malicious nodes, save the storage space of nodes greatly, and improve the efficiency of network operation. Moreover, SAM-PP has a further advantage in reducing the authentication delay.

A design-for-testability method and an electrical interconnect test method are proposed to detect open defects occurring at interconnects among dies and input/output pins in 3D stacked ICs. As part of the design method, an nMOS and a diode are added to each input interconnect. The test method is based on measuring the quiescent current that is made to flow through an interconnect to be tested. The testability is examined both by SPICE simulation and by experimentation. The test method enabled the detection of open defects occurring at the newly designed interconnects of dies at experiments test speed of 1MHz. The simulation results reveal that an open defect generating additional delay of 279psec is detectable by the test method at a test speed of 200MHz beside of open defects that generate no logical errors.

In this paper, a multilevel thresholding color image segmentation method is proposed using a modified Artificial Bee Colony(ABC) algorithm. In this work, in order to improve the local search ability of ABC algorithm, Krill Herd algorithm is incorporated into its onlooker bees phase. The proposed algorithm is named as Krill herd-inspired modified Artificial Bee Colony algorithm (KABC algorithm). Experiment results verify the robustness of KABC algorithm, as well as its improvement in optimizing accuracy and convergence speed. In this work, KABC algorithm is used to solve the problem of multilevel thresholding for color image segmentation. To deal with luminance variation, rather than using gray scale histogram, a HSV space-based pre-processing method is proposed to obtain 1D feature vector. KABC algorithm is then applied to find thresholds of the feature vector. At last, an additional local search around the quasi-optimal solutions is employed to improve segmentation accuracy. In this stage, we use a modified objective function which combines Structural Similarity Index Matrix (SSIM) with Kapur's entropy. The pre-processing method, the global optimization with KABC algorithm and the local optimization stage form the whole color image segmentation method. Experiment results show enhance in accuracy of segmentation with the proposed method.

In historical learning, to grasp the causal relationship between historical events and to understand factors that bring about important events are significant for fostering the historical thinking. However, some students are not able to find historical events that have causal relationships. The view of observing the historical events is different among individuals, so it is not appropriate to define the historical events that have causal relationships and impose students to remember them. The students need to understand the definition of the causal relationships and find the historical events that satisfy the definition according to their viewpoints. The objective of this paper is to develop a support system for understanding the meaning of a causal relationship and creating causal relation graphs that represent the causal relationships between historical events. When historical events have a causal relationship, a state change caused by one event becomes the cause of the other event. To consider these state changes is critically important to connect historical events. This paper proposes steps for considering causal relationships between historical events by arranging the state changes of historical people along with them. It also develops the system that supports students to create the causal relation graph according to the state changes. In our system, firstly, the interface for arranging state changes of historical people according to the historical events is given. Then, the interface for drawing the causal relation graph of historical events is provided in which state changes are automatically indicated on the created links in the causal relation graph. By observing the indicated state changes on the links, students are able to check by themselves whether their causal relation graphs correctly represent the causal relationships between historical events.

Objective: Detection of Electrocardiogram (ECG) characteristic points can provide critical diagnostic information about heart diseases. We proposed a novel feature extraction and machine learning scheme for automatic detection of ECG characteristic points. Methods: A new feature, termed as randomly selected wavelet transform (RSWT) feature, was devised to represent ECG characteristic points. A random forest classifier was adapted to infer the characteristic points position with high sensitivity and precision. Results: Compared with other state-of-the-art algorithms' testing results on QT database, our detection results of RSWT scheme showed comparable performance (similar sensitivity, precision, and detection error for each characteristic point). RSWT testing on MIT-BIH database also demonstrated promising cross-database performance. Conclusion: A novel RSWT feature and a new detection scheme was fabricated for ECG characteristic points. The RSWT demonstrated a robust and trustworthy feature for representing ECG morphologies. Significance: With the effectiveness of the proposed RSWT feature we presented a novel machine learning based scheme to automatically detect all types of ECG characteristic points at a time. Furthermore, it showed that our algorithm achieved better performance than other reported machine learning based methods.

As interaction between human and computer continues to develop to the most natural form possible, it becomes increasingly urgent to incorporate emotion in the equation. This paper describes a step toward extending the research on emotion recognition to Indonesian. The field continues to develop, yet exploration of the subject in Indonesian is still lacking. In particular, this paper highlights two contributions: (1) the construction of the first emotional audio-visual database in Indonesian, and (2) the first multimodal emotion recognizer in Indonesian, built from the aforementioned corpus. In constructing the corpus, we aim at natural emotions that are corresponding to real life occurrences. However, the collection of emotional corpora is notably labor intensive and expensive. To diminish the cost, we collect the emotional data from television programs recordings, eliminating the need of an elaborate recording set up and experienced participants. In particular, we choose television talk shows due to its natural conversational content, yielding spontaneous emotion occurrences. To cover a broad range of emotions, we collected three episodes in different genres: politics, humanity, and entertainment. In this paper, we report points of analysis of the data and annotations. The acquisition of the emotion corpus serves as a foundation in further research on emotion. Subsequently, in the experiment, we employ the support vector machine (SVM) algorithm to model the emotions in the collected data. We perform multimodal emotion recognition utilizing the predictions of three modalities: acoustic, semantic, and visual. When compared to the unimodal result, in the multimodal feature combination, we attain identical accuracy for the arousal at 92.6%, and a significant improvement for the valence classification task at 93.8%. We hope to continue this work and move towards a finer-grain, more precise quantification of emotion.

An improved radiometric calibration algorithm by extending the Mitsunaga and Nayar least-square minimization based algorithm with two major ideas is presented. First, a noise & outlier removal procedure based on the analysis of brightness transfer function is included for improving the algorithm's capability on handling noise and outlier in least-square estimation. Second, an alternative minimization formulation based on weighted least square is proposed to improve the weakness of least square minimization when dealing with biased distribution observations. The performance of the proposed algorithm with regards to two baseline algorithms is demonstrated, i.e. the classical least square based algorithm proposed by Mitsunaga and Nayar and the state-of-the-art rank minimization based algorithm proposed by Lee et al. From the results, the proposed algorithm outperforms both baseline algorithms on both the synthetic dataset and the dataset of real-world images.

Multi-task joint sparse representation (MTJSR) is one kind of efficient multi-task learning (MTL) method for solving different problems together using a shared sparse representation. Based on the learning mechanism in human, which is a self-paced learning by gradually training the tasks from easy to difficult, I apply this mechanism into MTJSR, and propose a multi-task joint sparse representation with self-paced learning (MTJSR-SP) algorithm. In MTJSR-SP, the self-paced learning mechanism is considered as a regularizer of optimization function, and an iterative optimization is applied to solve it. Comparing with the traditional MTL methods, MTJSR-SP has more robustness to the noise and outliers. The experimental results on some datasets, i.e. two synthesized datasets, four datasets from UCI machine learning repository, an oxford flower dataset and a Caltech-256 image categorization dataset, are used to validate the efficiency of MTJSR-SP.

Sparse representation has been widely applied to visual tracking for several years. In the sparse representation framework, tracking problem is transferred into solving an L1 minimization issue. However, during the tracking procedure, the appearance of target was affected by external environment. Therefore, we proposed a robust tracking algorithm based on the traditional sparse representation jointly particle filter framework. First, we obtained the observation image set from particle filter. Furthermore, we introduced a 2D transformation on the observation image set, which enables the tracking target candidates set more robust to handle misalignment problem in complex scene. Moreover, we adopt the occlusion detection mechanism before template updating, reducing the drift problem effectively. Experimental evaluations on five public challenging sequences, which exhibit occlusions, illuminating variations, scale changes, motion blur, and our tracker demonstrate accuracy and robustness in comparisons with the state-of-the-arts.

We propose a new signature-based, on-demand anti-virus solution using in-storage processing (ISP) to inspect the inside of a storage device. In-storage anti-virus systems are able to isolate malicious effects from main computing platforms, and they reduce the system overhead for virus detection. We implement our in-storage anti-virus platform using cost-effective, open-source hardware, and we verify that is practically applicable to storage devices.

RTA (Response time analysis) is a popular technique to guarantee timing requirements for a real-time system, and therefore the RTA framework has been widely studied for popular scheduling algorithms such as EDF (Earliest Deadline First) and FP (Fixed Priority). While a number of extended techniques of RTA have been introduced, some of them cannot be used since they have not been proved and evaluated in terms of their correctness and empirical performance. In this letter, we address the state of the art technique of slack reclamation of the existing generic RTA framework for multiprocessors. We present its mathematical proof of correctness and empirical performance evaluation, which have not been revealed to this day.

The traditional Markov prediction methods of the taxi destination rely only on the previous 2 to 3 GPS points. They negelect long-term dependencies within a taxi trajectory. We adopt a Recurrent Neural Network (RNN) to explore the long-term dependencies to predict the taxi destination as the multiple hidden layers of RNN can store these dependencies. However, the hidden layers of RNN are very sensitive to small perturbations to reduce the prediction accuracy when the amount of taxi trajectories is increasing. In order to improve the prediction accuracy of taxi destination and reduce the training time, we embed suprisal-driven zoneout (SDZ) to RNN, hence a taxi destination prediction method by regularized RNN with SDZ (TDPRS). SDZ can not only improve the robustness of TDPRS, but also reduce the training time by adopting partial update of parameters instead of a full update. Experiments with a Porto taxi trajectory data show that TDPRS improves the prediction accuracy by 12% compared to RNN prediction method in literature[4]. At the same time, the prediction time is reduced by 7%.

We have proposed a fish farm monitoring system for achieving efficient fish farming. In our system, sensor nodes are attached at fish to monitor its health status. In this letter, we propose a method for gathering sensor data from sensor nodes to sink nodes when the transmission range of sensor node is shorter than the size of fish cage. In our proposed method, a part of sensor nodes become leader nodes and they forward gathered sensor data to the sink nodes. Through simulation evaluations, we show that the data gathering performance of our proposed method is higher than that of traditional methods.

The rising systems programming language Rust is fast, efficient and memory safe. However, improperly dereferencing raw pointers in Rust causes new safety problems. In this paper, we present a detailed analysis into these problems and propose a practical hybrid approach to detecting unsafe raw pointer dereferencing behaviors. Our approach employs pattern matching to identify functions that can be used to generate illegal multiple mutable references (We define them as thief function) and instruments the dereferencing operation in order to perform dynamic checking at runtime. We implement a tool named UnsafeFencer and has successfully identified 52 thief functions in 28 real-world crates*, of which 13 public functions are verified to generate multiple mutable references.

To learn the working situation of rod-pumped wells under ground, we always need to analyze dynamometer diagrams, which are generated by the load sensor and displacement sensor. Rod-pumped wells are usually located in the places with extreme weather, and these sensors are installed on some special oil equipments in the open air. As time goes by, sensors are prone to generating unstable and incorrect data. Unfortunately, load sensors are too expensive to frequently reinstall. Therefore, the resulting dynamometer diagrams sometimes cannot make an accurate diagnosis. Instead, as an absolutely necessary equipment of the rod-pumped well, the electric motor has much longer life and cannot be easily impacted by the weather. The electric power curve during a swabbing period can also reflect the working situation under ground, but is much harder to explain than the dynamometer diagram. This letter presented a novel deep learning architecture, which can transform the electric power curve into the dimensionless dynamometer diagram image. We conduct our experiments on a real-world dataset, and the results show that our method can get an impressive transformation accuracy.

PCANet, as one noticeable shallow network, employs the histogram representation for feature pooling. However, there are three main problems about this kind of pooling method. First, the histogram-based pooling method binarizes the feature maps and leads to inevitable discriminative information loss. Second, it is difficult to effectively combine other visual cues into a compact representation, because the simple concatenation of various visual cues leads to feature representation inefficiency. Third, the dimensionality of histogram-based output grows exponentially with the number of feature maps used. In order to overcome these problems, we propose a novel shallow network model, named as PCANet-II. Compared with the histogram-based output, the second order pooling not only provides more discriminative information by preserving both the magnitude and sign of convolutional responses, but also dramatically reduces the size of output features. Thus we combine the second order statistical pooling method with the shallow network, i.e., PCANet. Moreover, it is easy to combine other discriminative and robust cues by using the second order pooling. So we introduce the binary feature difference encoding scheme into our PCANet-II to further improve robustness. Experiments demonstrate the effectiveness and robustness of our proposed PCANet-II method.

This paper presents an adaptive least-significant-bit (LSB) steganography for spatial color images on smartphones. For each red, green, and blue color component, the combinations of All-4bit, One-4bit+Two-2bit, and Two-3bit+One-2bit LSB replacements are proposed for content-adaptivity and natural histograms. The high capacity of 8.4bpp with the average peak signal noise ratio (PSNR) 43.7db and fast processing times on smartphones are also demonstrated

We present a system to improve the robustness of real-time 3D surface reconstruction by utilizing non-inertial localization sensor. Benefiting from such sensor, our easy-to-build system can effectively avoid tracking drift and lost comparing with conventional dense tracking and mapping systems. To best fusing the sensor, we first adopt a hand-eye calibration and performance analysis for our setup and then propose a novel optimization framework based on adaptive criterion function to improve the robustness as well as accuracy. We apply our system to several challenging reconstruction tasks, which show significant improvement in scanning robustness and reconstruction quality.

We propose a novel neural networks model based on LSTM which is used to solve the task of classifying inertial sensor data attached to a fence with the goal of detecting security relevant incidents. To evaluate it we deployed an experimental fence surveillance system. By comparing experimental data of different approaches we find out that the neural network outperforms the baseline approach.