IEICE Transactions on Information and Systems
Online ISSN : 1745-1361
Print ISSN : 0916-8532
Volume E101.D , Issue 4
Showing 1-44 articles out of 44 articles from the selected issue
Special Section on Intelligent Information and Communication Technology and its Applications to Creative Activity Support
  • Susumu KUNIFUJI, Thanaruk THEERAMUNKONG
    2018 Volume E101.D Issue 4 Pages 836-837
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS
    Download PDF (170K)
  • Jun MUNEMORI, Hiroki SAKAMOTO, Junko ITOU
    Type: PAPER
    Subject area: Creativity Support Systems and Decision Support Systems
    2018 Volume E101.D Issue 4 Pages 838-846
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In recent years, networking has spread substantially owing to the rapid developments made in Information & Communication Technology (ICT). It has also become easy to share highly contextual data and information, including ideas, among people. On the other hand, there exists information that cannot be expressed in words (tacit knowledge) and useful knowledge or know-how that is not shared well in an organization. The idea generation method enables the expression of explicit knowledge, which enables the expression of tacit knowledge by words, and can utilize explicit knowledge as know-how in organizations. We propose an idea generation consistent support system, GUNGEN-Web II. This system has suggestion functions for a concrete idea label and a concrete island name. The suggestion functions convey an idea and the island name to other participants more precisely. This system also has an illustration support function and a document support function. In this study, we aimed to improve the quality of the sentence obtained using the KJ method. We compared the results of our proposed systems with conventional GUNGEN-Web by conducting experiments. The results are as follows: The evaluation of the sentence of GUNGEN-Web II was significantly different to those obtained using the conventional GUNGEN-Web.

    Download PDF (1322K)
  • Satoshi KAWASE, Takayuki ITO, Takanobu OTSUKA, Akihisa SENGOKU, Shun S ...
    Type: PAPER
    Subject area: Creativity Support Systems and Decision Support Systems
    2018 Volume E101.D Issue 4 Pages 847-855
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Performance based on multi-party discussion has been reported to be superior to that based on individuals. However, it is impossible that all participants simultaneously express opinions due to the time and space limitations in a large-scale discussion. In particular, only a few representative discussants and audiences can speak in conventional unidirectional discussions (e.g., panel discussion), although many participants gather for the discussion. To solve these problems, in this study, we proposed a cyber-physical discussion using “COLLAGREE,” which we developed for building consensus of large-scale online discussions. COLLAGREE is equipped with functions such as a facilitator, point ranking system, and display of discussion in tree structure. We focused on the relationship between satisfaction with the discussion and participants' desire to express opinions. We conducted the experiment in the panel discussion of an actual international conference. Participants who were audiences in the floor used COLLAGREE during the panel discussion. They responded to questionnaires after the experiment. The main findings are as follows: (1) Participation in online discussion was associated with the satisfaction of the participants; (2) Participants who desired to positively express opinions joined the cyber-space discussion; and (3) The satisfaction of participants who expressed opinions in the cyber-space discussion was higher than those of participants who expressed opinions in the real-space discussion and those who did not express opinions in both the cyber- and real-space discussions. Overall, active behaviors in the cyber-space discussion were associated with participants' satisfaction with the entire discussion, suggesting that cyberspace provided useful alternative opportunities to express opinions for audiences who used to listen to conventional unidirectional discussions passively. In addition, a complementary relationship exists between participation in the cyber-space and real-space discussions. These findings can serve to create a user-friendly discussion environment.

    Download PDF (1554K)
  • Shota KUSAJIMA, Yasuyuki SUMI
    Type: PAPER
    Subject area: Creativity Support Systems and Decision Support Systems
    2018 Volume E101.D Issue 4 Pages 856-864
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Online chat systems, e.g.., Twitter and Slack, have been used in academic conferences or study meetings as a means of instant discussion and sharing related information alongside a real presentation. We propose a system for activating online discussion by providing a bot that suggests webpages related to current timeline of the discussion. Our system generates keyword vectors according to discussion timeline, searches best related webpages from several web sites, and timely provides these pages to the discussion timeline. This paper describes deployments of our system in two types of meetings: lightning talk format meetings and group meetings; and daily exchanges using online chat system. As a result, we could not find good enough reactions to the bot's postings from meeting participants at the lightning talk format meetings, but we could observe more reactions and progress of discussion caused by the bot's postings at the relaxed meetings and daily exchanges among group members.

    Download PDF (1763K)
  • Katsuhide FUJITA, Ryosuke WATANABE
    Type: PAPER
    Subject area: Creativity Support Systems and Decision Support Systems
    2018 Volume E101.D Issue 4 Pages 865-873
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Recently, the opportunity to discuss topics on a variety of online discussion bulletin boards has been increasing. However, it can be difficult to understand the contents of each discussion as the number of posts increases. Therefore, it is important to generate headlines that can automatically summarize each post in order to understand the contents of each discussion at a glance. In this paper, we propose a method to extract and generate post headlines for online discussion bulletin boards, automatically. We propose templates with multiple patterns to extract important sentences from the posts. In addition, we propose a method to generate headlines by matching the templates with the patterns. Then, we evaluate the effectiveness of our proposed method using questionnaires.

    Download PDF (391K)
  • Hiroyuki MITSUHARA, Masami SHISHIBORI, Akihiro KASHIHARA
    Type: PAPER
    Subject area: Creativity Support Systems and Decision Support Systems
    2018 Volume E101.D Issue 4 Pages 874-883
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Investigative reports plagiarized from the web should be eliminated because such reports result in ineffective knowledge construction. In this study, we developed an investigative report writing support system for effective knowledge construction from the web. The proposed system attempts to prevent plagiarism by restricting copying and pasting information from web pages. With this system, students can verify information through web browsing, externalize their constructed knowledge as notes for report materials, write reports using these notes, and remove inadequacies in the report by reflection. A comparative experiment showed that the proposed system can potentially prevent web page plagiarism and make knowledge construction from the web more effective compared to a conventional report writing environment.

    Download PDF (1221K)
  • Marut BURANARACH, Chutiporn ANUTARIYA, Nopachat KALAYANAPAN, Taneth RU ...
    Type: PAPER
    Subject area: Knowledge Representation
    2018 Volume E101.D Issue 4 Pages 884-891
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Knowledge management is important for government agencies in improving service delivery to their customers and data inter-operation within and across organizations. Building organizational knowledge repository for government agency has unique challenges. In this paper, we propose that enterprise ontology can provide support for government agencies in capturing organizational taxonomy, best practices and global data schema. A case study of a large-scale adoption for the Thailand's Excise Department is elaborated. A modular design approach of the enterprise ontology for the excise tax domain is discussed. Two forms of organizational knowledge: global schema and standard practices were captured in form of ontology and rule-based knowledge. The organizational knowledge was deployed to support two KM systems: excise recommender service and linked open data. Finally, we discuss some lessons learned in adopting the framework in the government agency.

    Download PDF (1467K)
  • Akkharawoot TAKHOM, Sasiporn USANAVASIN, Thepchai SUPNITHI, Mitsuru IK ...
    Type: PAPER
    Subject area: Knowledge Representation
    2018 Volume E101.D Issue 4 Pages 892-900
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Creating an ontology from multidisciplinary knowledge is a challenge because it needs a number of various domain experts to collaborate in knowledge construction and verify the semantic meanings of the cross-domain concepts. Confusions and misinterpretations of concepts during knowledge creation are usually caused by having different perspectives and different business goals from different domain experts. In this paper, we propose a community-driven ontology-based application management (CD-OAM) framework that provides a collaborative environment with supporting features to enable collaborative knowledge creation. It can also reduce confusions and misinterpretations among domain stakeholders during knowledge construction process. We selected one of the multidisciplinary domains, which is Life Cycle Assessment (LCA) for our scenario-based knowledge construction. Constructing the LCA knowledge requires many concepts from various fields including environment protection, economic development, social development, etc. The output of this collaborative knowledge construction is called MLCA (multidisciplinary LCA) ontology. Based on our scenario-based experiment, it shows that CD-OAM framework can support the collaborative activities for MLCA knowledge construction and also reduce confusions and misinterpretations of cross-domain concepts that usually presents in general approach.

    Download PDF (3484K)
  • Takashi MATSUBARA, Ryo AKITA, Kuniaki UEHARA
    Type: PAPER
    Subject area: Datamining Technologies
    2018 Volume E101.D Issue 4 Pages 901-908
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In this study, we propose a deep neural generative model for predicting daily stock price movements given news articles. Approaches involving conventional technical analysis have been investigated to identify certain patterns in past price movements, which in turn helps to predict future price movements. However, the financial market is highly sensitive to specific events, including corporate buyouts, product releases, and the like. Therefore, recent research has focused on modeling relationships between these events that appear in the news articles and future price movements; however, a very large number of news articles are published daily, each article containing rich information, which results in overfitting to past price movements used for parameter adjustment. Given the above, we propose a model based on a generative model of news articles that includes price movement as a condition, thereby avoiding excessive overfitting thanks to the nature of the generative model. We evaluate our proposed model using historical price movements of Nikkei 225 and Standard & Poor's 500 Stock Index, confirming that our model predicts future price movements better than such conventional classifiers as support vector machines and multilayer perceptrons. Further, our proposed model extracts significant words from news articles that are directly related to future stock price movements.

    Download PDF (406K)
  • Tran Sy BANG, Virach SORNLERTLAMVANICH
    Type: PAPER
    Subject area: Datamining Technologies
    2018 Volume E101.D Issue 4 Pages 909-916
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper presents a supervised method to classify a document at the sub-sentence level. Traditionally, sentiment analysis often classifies sentence polarity based on word features, syllable features, or N-gram features. A sentence, as a whole, may contain several phrases and words which carry their own specific sentiment. However, classifying a sentence based on phrases and words can sometimes be incoherent because they are ungrammatically formed. In order to overcome this problem, we need to arrange words and phrase in a dependency form to capture their semantic scope of sentiment. Thus, we transform a sentence into a dependency tree structure. A dependency tree is composed of subtrees, and each subtree allocates words and syllables in a grammatical order. Moreover, a sentence dependency tree structure can mitigate word sense ambiguity or solve the inherent polysemy of words by determining their word sense. In our experiment, we provide the details of the proposed subtree polarity classification for sub-opinion analysis. To conclude our discussion, we also elaborate on the effectiveness of the analysis result.

    Download PDF (1886K)
  • Tessai HAYAMA
    Type: PAPER
    Subject area: Datamining Technologies
    2018 Volume E101.D Issue 4 Pages 917-924
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper presents a novel TV event detection method for automatically generating TV program digests by using Twitter data. Previous studies of TV program digest generation based on Twitter data have developed TV event detection methods that analyze the frequency time series of tweets that users made while watching a given TV program; however, in most of the previous studies, differences in how Twitter is used, e.g., sharing information versus conversing, have not been taken into consideration. Since these different types of Twitter data are lumped together into one category, it is difficult to detect highlight scenes of TV programs and correctly extract their content from the Twitter data. Therefore, this paper presents a highlight scene detection method to automatically generate TV program digests for TV programs based on Twitter data classified by Twitter user behavior. To confirm the effectiveness of the proposed method, experiments using 49 soccer game TV programs were conducted.

    Download PDF (1533K)
  • Maoshen JIA, Jundai SUN, Feng DENG, Junyue SUN
    Type: PAPER
    Subject area: Elemental Technologies for human behavior analysis
    2018 Volume E101.D Issue 4 Pages 925-932
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In this work, a multiple source separation method with joint sparse and non-sparse components recovery is proposed by using dual similarity determination. Specifically, a dual similarity coefficient is designed based on normalized cross-correlation and Jaccard coefficients, and its reasonability is validated via a statistical analysis on a quantitative effective measure. Thereafter, by regarding the sparse components as a guide, the non-sparse components are recovered using the dual similarity coefficient. Eventually, a separated signal is obtained by a synthesis of the sparse and non-sparse components. Experimental results demonstrate the separation quality of the proposed method outperforms some existing BSS methods including sparse components separation based methods, independent components analysis based methods and soft threshold based methods.

    Download PDF (644K)
  • Hiroki CHIBA, Yuki HYOGO, Kazuo MISUE
    Type: PAPER
    Subject area: Elemental Technologies for human behavior analysis
    2018 Volume E101.D Issue 4 Pages 933-943
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Spatio-temporal dependent data, such as weather observation data, are data of which the attribute values depend on both time and space. Typical methods for the visualization of such data include plotting the attribute values at each point in time on a map and displaying series of the maps in chronological order with animation, or displaying them by juxtaposing horizontally or vertically. However, these methods are problematic in that they compel readers interested in grasping the spatial changes of the attribute values to memorize the representations on the maps. The problem is exacerbated by considering that the longer the time-period covered by the data, the higher the cognitive load. In order to solve these problems, the authors propose a visualization method capable of overlaying the representations of multiple instantaneous values on a single static map. This paper explains the design of the proposed method and reports two experiments conducted by the authors to investigate the usefulness of the method. The experimental results show that the proposed method is useful in terms of the speed and accuracy with which it reads the spatial changes and its ability to present data with long time series efficiently.

    Download PDF (2825K)
  • Yasuhiro MOCHIDA, Daisuke SHIRAI, Tatsuya FUJII
    Type: PAPER
    Subject area: Technologies for Knowledge Support Platform
    2018 Volume E101.D Issue 4 Pages 944-955
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Existing remote collaboration systems are not suitable for a collaboration style where distributed users touch work tools at the same time, especially in demanding use cases or in severe network situations. To cover a wider range of use cases, we propose a novel concept of a remote collaboration platform that enables the users to share currently-used work tools with a high quality A/V transmission module, while maintaining the advantages of web-based systems. It also provides functions to deal with long transmission delay using relay servers, packet transmission instability using visual feedback of audio delivery and limited bandwidth using dynamic allocation of video bitrate. We implemented the platform and conducted evaluation tests. The results show the feasibility of the proposed concept and its tolerance to network constraints, which indicates that the proposed platform can construct unprecedented collaboration systems.

    Download PDF (2009K)
  • Juan YU, Peizhong LU, Jianmin HAN, Jianfeng LU
    Type: PAPER
    Subject area: Technologies for Knowledge Support Platform
    2018 Volume E101.D Issue 4 Pages 956-963
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Traffic signal phase and timing (TSPaT) information is valuable for various applications, such as velocity advisory systems, navigation systems, collision warning systems, and so forth. In this paper, we focus on learning baseline timing cycle lengths for fixed-time traffic signals. The cycle length is the most important parameter among all timing parameters, such as green lengths. We formulate the cycle length learning problem as a period estimation problem using a sparse set of noisy observations, and propose the most frequent approximate greatest common divisor (MFAGCD) algorithms to solve the problem. The accuracy performance of our proposed algorithms is experimentally evaluated on both simulation data and the real taxi GPS trajectory data collected in Shanghai, China. Experimental results show that the MFAGCD algorithms have better sparsity and outliers tolerant capabilities than existing cycle length estimation algorithms.

    Download PDF (5967K)
  • Miyoung JANG, Jae-Woo CHANG
    Type: PAPER
    Subject area: Technologies for Knowledge Support Platform
    2018 Volume E101.D Issue 4 Pages 964-976
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Recently, the join processing of large-scale datasets in MapReduce environments has become an important issue. However, the existing MapReduce-based join algorithms suffer from too much overhead for constructing and updating the data index. Moreover, the similarity computation cost is high because the existing algorithms partition data without considering the data distribution. In this paper, we propose two grid-based join algorithms for MapReduce. First, we propose a similarity join algorithm that evenly distributes join candidates using a dynamic grid index, which partitions data considering data density and similarity threshold. We use a bottom-up approach by merging initial grid cells into partitions and assigning them to MapReduce jobs. Second, we propose a k-NN join query processing algorithm for MapReduce. To reduce the data transmission cost, we determine an optimal grid cell size by considering the data distribution of randomly selected samples. Then, we perform kNN join by assigning the only related join data to a reducer. From performance analysis, we show that our similarity join query processing algorithm and our k-NN join algorithm outperform existing algorithms by up to 10 times, in terms of query processing time.

    Download PDF (2579K)
  • Ying-Yao TING, Chi-Wei HSIAO, Huan-Sheng WANG
    Type: PAPER
    Subject area: Technologies for Knowledge Support Platform
    2018 Volume E101.D Issue 4 Pages 977-984
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    To prevent constraints or defects of a single sensor from malfunctions, this paper proposes a fire detection system based on the Dempster-Shafer theory with multi-sensor technology. The proposed system operates in three stages: measurement, data reception and alarm activation, where an Arduino is tasked with measuring and interpreting the readings from three types of sensors. Sensors under consideration involve smoke, light and temperature detection. All the measured data are wirelessly transmitted to the backend Raspberry Pi for subsequent processing. Within the system, the Raspberry Pi is used to determine the probability of fire events using the Dempster-Shafer theory. We investigate moderate settings of the conflict coefficient and how it plays an essential role in ensuring the plausibility of the system's deduced results. Furthermore, a MySQL database with a web server is deployed on the Raspberry Pi for backlog and data analysis purposes. In addition, the system provides three notification services, including web browsing, smartphone APP, and short message service. For validation, we collected the statistics from field tests conducted in a controllable and safe environment by emulating fire events happening during both daytime and nighttime. Each experiment undergoes the No-fire, On-fire and Post-fire phases. Experimental results show an accuracy of up to 98% in both the No-fire and On-fire phases during the daytime and an accuracy of 97% during the nighttime under reasonable conditions. When we take the three phases into account, the accuracy in the daytime and nighttime increase to 97% and 89%, respectively. Field tests validate the efficiency and accuracy of the proposed system.

    Download PDF (1311K)
Special Section on Data Engineering and Information Management
  • Masashi TOYODA
    2018 Volume E101.D Issue 4 Pages 985
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS
    Download PDF (83K)
  • Ponrudee NETISOPAKUL, Gerhard WOHLGENANNT
    Type: SURVEY PAPER
    2018 Volume E101.D Issue 4 Pages 986-1002
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    As the manual creation of domain models and also of linked data is very costly, the extraction of knowledge from structured and unstructured data has been one of the central research areas in the Semantic Web field in the last two decades. Here, we look specifically at the extraction of formalized knowledge from natural language text, which is the most abundant source of human knowledge available. There are many tools on hand for information and knowledge extraction for English natural language, for written Thai language the situation is different. The goal of this work is to assess the state-of-the-art of research on formal knowledge extraction specifically from Thai language text, and then give suggestions and practical research ideas on how to improve the state-of-the-art. To address the goal, first we distinguish nine knowledge extraction for the Semantic Web tasks defined in literature on knowledge extraction from English text, for example taxonomy extraction, relation extraction, or named entity recognition. For each of the nine tasks, we analyze the publications and tools available for Thai text in the form of a comprehensive literature survey. Additionally to our assessment, we measure the self-assessment by the Thai research community with the help of a questionnaire-based survey on each of the tasks. Furthermore, the structure and size of the Thai community is analyzed using complex literature database queries. Combining all the collected information we finally identify research gaps in knowledge extraction from Thai language. An extensive list of practical research ideas is presented, focusing on concrete suggestions for every knowledge extraction task - which can be implemented and evaluated with reasonable effort. Besides the task-specific hints for improvements of the state-of-the-art, we also include general recommendations on how to raise the efficiency of the respective research community.

    Download PDF (1081K)
  • Yasuhito ASANO, Junpei KAWAMOTO
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1003-1011
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Early reviews, posted on online review sites shortly after products enter the market, are useful for estimating long-term evaluations of those products and making decisions. However, such reviews can be influenced easily by anomalous reviewers, including malicious and fraudulent reviewers, because the number of early reviews is usually small. It is therefore challenging to detect anomalous reviewers from early reviews and estimate long-term evaluations by reducing their influences. We find that two characteristics of heterogeneity on actual review sites such as Amazon.com cause difficulty in detecting anomalous reviewers from early reviews. We propose ideas for consideration of heterogeneity, and a methodology for computing reviewers' degree of anomaly and estimating long-term evaluations simultaneously. Our experimental evaluations with actual reviews from Amazon.com revealed that our proposed method achieves the best performance in 19 of 20 tests compared to state-of-the-art methodologies.

    Download PDF (340K)
  • Yuyang DONG, Hanxiong CHEN, Kazutaka FURUSE, Hiroyuki KITAGAWA
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1012-1020
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Given two data sets of user preferences and product attributes in addition to a set of query products, the aggregate reverse rank (ARR) query returns top-k users who regard the given query products as the highest aggregate rank than other users. ARR queries are designed to focus on product bundling in marketing. Manufacturers are mostly willing to bundle several products together for the purpose of maximizing benefits or inventory liquidation. This naturally leads to an increase in data on users and products. Thus, the problem of efficiently processing ARR queries become a big issue. In this paper, we reveal two limitations of the state-of-the-art solution to ARR query; that is, (a) It has poor efficiency when the distribution of the query set is dispersive. (b) It has to process a large portion user data. To address these limitations, we develop a cluster-and-process method and a sophisticated indexing strategy. From the theoretical analysis of the results and experimental comparisons, we conclude that our proposals have superior performance.

    Download PDF (1875K)
  • Sopheaktra YONG, Yasuhito ASANO
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1021-1029
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    To help with decision making, online shoppers tend to go through both a list of a product's features and functionality provided by the vendor, as well as a list of reviews written by other users. Unfortunately, this process is ineffective when the buyer is confronted with large amounts of information, particularly when the buyer has limited experience with and knowledge of the product. In order to avoid this problem, we propose a framework of purpose-oriented recommendation that presents a ranked list of products suitable for a designated user purpose by identifying important product features to fulfill the purpose from online reviews. As technical foundation for realizing the framework, we propose several methods to mine relation between user purposes and product features from the consumer reviews. Using digital camera reviews on Amazon.com, the experimental results show that our proposed method is both effective and stable, with an acceptable rate of precision and recall.

    Download PDF (397K)
  • Akio WATANABE, Keisuke ISHIBASHI, Tsuyoshi TOYONO, Keishiro WATANABE, ...
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1030-1041
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In current large-scale IT systems, troubleshooting has become more complicated due to the diversification in the causes of failures, which has increased operational costs. Thus, clarifying the troubleshooting process also becomes important, though it is also time-consuming. We propose a method of automatically extracting a workflow, a graph indicating a troubleshooting process, using multiple trouble tickets. Our method extracts an operator's actions from free-format texts and aligns relative sentences between multiple trouble tickets. Our method uses a stochastic model to detect a resolution, a frequent action pattern that helps us understand how to solve a problem. We validated our method using real trouble-ticket data captured from a real network operation and showed that it can extract a workflow to identify the cause of a failure.

    Download PDF (1473K)
  • Ayae ICHINOSE, Atsuko TAKEFUSA, Hidemoto NAKADA, Masato OGUCHI
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1042-1052
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Many life-log analysis applications, which transfer data from cameras and sensors to a Cloud and analyze them in the Cloud, have been developed as the use of various sensors and Cloud computing technologies has spread. However, difficulties arise because of the limited network bandwidth between such sensors and the Cloud. In addition, sending raw sensor data to a Cloud may introduce privacy issues. Therefore, we propose a pipelined method for distributed deep learning processing between sensors and the Cloud to reduce the amount of data sent to the Cloud and protect the privacy of users. In this study, we measured the processing times and evaluated the performance of our method using two different datasets. In addition, we performed experiments using three types of machines with different performance characteristics on the client side and compared the processing times. The experimental results show that the accuracy of deep learning with coarse-grained data is comparable to that achieved with the default parameter settings, and the proposed distributed processing method has performance advantages in cases of insufficient network bandwidth between realistic sensors and a Cloud environment. In addition, it is confirmed that the process that most affects the overall processing time varies depending on the machine performance on the client side, and the most efficient distribution method similarly differs.

    Download PDF (3453K)
  • Hong Van LE, Atsuhiro TAKASU
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1053-1065
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    With the recent explosion of geographic data generated by smartphones, sensors, and satellites, a data storage that can handle the massive volume of data and support high-computational spatial queries is becoming essential. Although key-value stores efficiently handle large-scale data, they are not equipped with effective functions for supporting geographic data. To solve this problem, in this paper, we present G-HBase, a high-performance geographical database based on HBase, a standard key-value store. To index geographic data, we first use Geohash as the rowkey in HBase. Then, we present a novel partitioning method, namely binary Geohash rectangle partitioning, to support spatial queries. Our extensive experiments on real datasets have demonstrated an improved performance with k nearest neighbors and range query in G-HBase when compared with SpatialHadoop, a state-of-the-art framework with native support for spatial data. We also observed that performance of spatial join in G-HBase is on par with SpatialHadoop and outperforms SJMR algorithm in HBase.

    Download PDF (2385K)
  • Ikuo KESHI, Yu SUZUKI, Koichiro YOSHINO, Satoshi NAKAMURA
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1066-1078
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    The problem with distributed representations generated by neural networks is that the meaning of the features is difficult to understand. We propose a new method that gives a specific meaning to each node of a hidden layer by introducing a manually created word semantic vector dictionary into the initial weights and by using paragraph vector models. We conducted experiments to test the hypotheses using a single domain benchmark for Japanese Twitter sentiment analysis and then evaluated the expandability of the method using a diverse and large-scale benchmark. Moreover, we tested the domain-independence of the method using a Wikipedia corpus. Our experimental results demonstrated that the learned vector is better than the performance of the existing paragraph vector in the evaluation of the Twitter sentiment analysis task using the single domain benchmark. Also, we determined the readability of document embeddings, which means distributed representations of documents, in a user test. The definition of readability in this paper is that people can understand the meaning of large weighted features of distributed representations. A total of 52.4% of the top five weighted hidden nodes were related to tweets where one of the paragraph vector models learned the document embeddings. For the expandability evaluation of the method, we improved the dictionary based on the results of the hypothesis test and examined the relationship of the readability of learned word vectors and the task accuracy of Twitter sentiment analysis using the diverse and large-scale benchmark. We also conducted a word similarity task using the Wikipedia corpus to test the domain-independence of the method. We found the expandability results of the method are better than or comparable to the performance of the paragraph vector. Also, the objective and subjective evaluation support each hidden node maintaining a specific meaning. Thus, the proposed method succeeded in improving readability.

    Download PDF (1998K)
  • Jianfei XUE, Koji EGUCHI
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1079-1087
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Topic modeling as a well-known method is widely applied for not only text data mining but also multimedia data analysis such as video data analysis. However, existing models cannot adequately handle time dependency and multimodal data modeling for video data that generally contain image information and speech information. In this paper, we therefore propose a novel topic model, sequential symmetric correspondence hierarchical Dirichlet processes (Seq-Sym-cHDP) extended from sequential conditionally independent hierarchical Dirichlet processes (Seq-CI-HDP) and sequential correspondence hierarchical Dirichlet processes (Seq-cHDP), to improve the multimodal data modeling mechanism via controlling the pivot assignments with a latent variable. An inference scheme for Seq-Sym-cHDP based on a posterior representation sampler is also developed in this work. We finally demonstrate that our model outperforms other baseline models via experiments.

    Download PDF (1807K)
  • Xiaoying TAN, Yuchun GUO, Yishuai CHEN, Wei ZHU
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1088-1095
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    The Collaborative Filtering (CF) algorithms work fairly well in personalized recommendation except in sparse data environment. To deal with the sparsity problem, researchers either take into account auxiliary information extracted from additional data resources, or set the missing ratings with default values, e.g., video popularity. Nevertheless, the former often costs high and incurs difficulty in knowledge transference whereas the latter degrades the accuracy and coverage of recommendation results. To our best knowledge, few literatures take advantage of users' preference on video popularity to tackle this problem. In this paper, we intend to enhance the performance of recommendation algorithm via the inference of the users' popularity preferences (PPs), especially in a sparse data environment. We propose a scheme to aggregate users' PPs and a Collaborative Filtering based algorithm to make the inference of PP feasible and effective from a small number of watching records. We modify a k-Nearest-Neighbor recommendation algorithm and a Matrix Factorization algorithm via introducing the inferred PP. Experiments on a large-scale commercial dataset show that the modified algorithm outperforms the original CF algorithms on both the recommendation accuracy and coverage. The significance of improvement is significant especially with the data sparsity.

    Download PDF (630K)
  • Kosetsu TSUKUDA, Keisuke ISHIDA, Masahiro HAMASAKI, Masataka GOTO
    Type: PAPER
    2018 Volume E101.D Issue 4 Pages 1096-1106
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Creating new content based on existing original work is becoming popular especially among amateur creators. Such new content is called derivative work and can be transformed into the next new derivative work. Such derivative work creation is called “N-th order derivative creation.” Although derivative creation is popular, the reason an individual derivative work was created is not observable. To infer the factors that trigger derivative work creation, we have proposed a model that incorporates three factors: (1) original work's attractiveness, (2) original work's popularity, and (3) derivative work's popularity. Based on this model, in this paper, we describe a public web service for browsing derivation factors called Songrium Derivation Factor Analysis. Our service is implemented by applying our model to original works and derivative works uploaded to a video sharing service. Songrium Derivation Factor Analysis provides various visualization functions: Original Works Map, Derivation Tree, Popularity Influence Transition Graph, Creator Distribution Map, and Creator Profile. By displaying such information when users browse and watch videos, we aim to enable them to find new content and understand the N-th order derivative creation activity at a deeper level.

    Download PDF (3738K)
Regular Section
  • Shinji KAWAMURA, Tomoaki TSUMURA
    Type: PAPER
    Subject area: Computer System
    2018 Volume E101.D Issue 4 Pages 1107-1115
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Many mobile systems need to achieve both high performance and low memory usage, and the total performance of such the systems can be largely affected by the effectiveness of GC. Hence, the recent popularization of mobile devices makes the GC performance play one of the important roles on the wide range of platforms. The response performance degradation caused by suspending all processes for GC has been a well-known potential problem. Therefore, GC algorithms have been actively studied and improved, but they still have not reached any fundamental solution. In this paper, we focus on the point that the same objects are redundantly marked during the GC procedure implemented on DalvikVM, which is one of the famous runtime environments for the mobile devices. Then we propose a hardware support technique for improving marking routine of GC. We installed a set of tables to a processor for managing marked objects, and redundant marking for marked objects can be omitted by referring these tables. The result of the simulation experiment shows that the percentage of redundant marking is reduced by more than 50%.

    Download PDF (1376K)
  • Carlos Cesar CORTES TORRES, Hayate OKUHARA, Nobuyuki YAMASAKI, Hidehar ...
    Type: PAPER
    Subject area: Computer System
    2018 Volume E101.D Issue 4 Pages 1116-1125
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In the past decade, real-time systems (RTSs), which must maintain time constraints to avoid catastrophic consequences, have been widely introduced into various embedded systems and Internet of Things (IoTs). The RTSs are required to be energy efficient as they are used in embedded devices in which battery life is important. In this study, we investigated the RTS energy efficiency by analyzing the ability of body bias (BB) in providing a satisfying tradeoff between performance and energy. We propose a practical and realistic model that includes the BB energy and timing overhead in addition to idle region analysis. This study was conducted using accurate parameters extracted from a real chip using silicon on thin box (SOTB) technology. By using the BB control based on the proposed model, about 34% energy reduction was achieved.

    Download PDF (1417K)
  • Alimujiang YASEN, Kazunori UEDA
    Type: PAPER
    Subject area: Software System
    2018 Volume E101.D Issue 4 Pages 1126-1140
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    We develop a technique for representing variable names and name binding which is a mechanism of associating a name with an entity in many formal systems including logic, programming languages and mathematics. The idea is to use a general form of graph links (or edges) called hyperlinks to represent variables, graph nodes as constructors of the formal systems, and a graph type called hlground to define substitutions. Our technique is based on simple notions of graph theory in which graph types ensure correct substitutions and keep bound variables distinct. We encode strong reduction of the untyped λ-calculus to introduce our technique. Then we encode a more complex formal system called System F<:, a polymorphic λ-calculus with subtyping that has been one of important theoretical foundations of functional programming languages. The advantage of our technique is that the representation of terms, definition of substitutions, and implementation of formal systems are all straightforward. We formalized the graph type hlground, proved that it ensures correct substitutions in the λ-calculus, and implemented hlground in HyperLMNtal, a modeling language based on hypergraph rewriting. Experiments were conducted to test this technique. By this technique, one can implement formal systems simply by following the steps of their definitions as described in papers.

    Download PDF (989K)
  • Warunya WUNNASRI, Jaruwat PAILAI, Yusuke HAYASHI, Tsukasa HIRASHIMA
    Type: PAPER
    Subject area: Educational Technology
    2018 Volume E101.D Issue 4 Pages 1141-1150
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper describes an investigation into the validity of an automatic assessment method of the learner-build concept map by comparing it with two well-known manual methods. We have previously proposed the Kit-Build (KB) concept map framework where a learner builds a concept map by using only a provided set of components, known as the set “kit”. In this framework, instant and automatic assessment of a learner-build concept map has been realized. We call this assessment method the “Kit-Build method” (KB method). The framework and assessment method have already been practically used in classrooms in various schools. As an investigation of the validity of this method, we have conducted an experiment as a case study to compare the assessment results of the method with the assessment results of two other manual assessment methods. In this experiment, 22 university students attended as subjects and four as raters. It was found that the scores of the KB method had a very strong correlation with the scores of the other manual methods. The results of this experiment are one of evidence to show the automatic assessment of the Kit-Build concept map can attain almost the same level of validity as well-known manual assessment methods.

    Download PDF (1563K)
  • Chihiro TSUTAKE, Toshiyuki YOSHIDA
    Type: PAPER
    Subject area: Image Processing and Video Processing
    2018 Volume E101.D Issue 4 Pages 1151-1158
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Many of affine motion compensation techniques proposed thus far employ least-square-based techniques in estimating affine parameters, which requires a hardware structure different from conventional block-matching-based one. This paper proposes a new affine motion estimation/compensation framework friendly to block-matching-based parameter estimation, and applies it to an HEVC encoder to demonstrate its coding efficiency and computation cost. To avoid a nest of search loops, a new affine motion model is first introduced by decomposing the conventional 4-parameter affine model into two 3-parameter ones. Then, a block-matching-based fast parameter estimation technique is proposed for the models. The experimental results given in this paper show that our approach is advantageous over conventional techniques.

    Download PDF (414K)
  • Yinghui ZHANG, Hongjun WANG, Hengxue ZHOU, Ping DENG
    Type: PAPER
    Subject area: Image Processing and Video Processing
    2018 Volume E101.D Issue 4 Pages 1159-1166
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Image boundary detection or image segmentation is an important step in image analysis. However, choosing appropriate parameters for boundary detection algorithms is necessary to achieve good boundary detection results. Image boundary detection fusion with unsupervised parameters can output a final consensus boundary, which is generally better than using unsupervised or supervised image boundary detection algorithms. In this study, we theoretically examine why image boundary detection fusion can work well and we propose a mixture model for image boundary detection fusion (MMIBDF) to achieve good consensus segmentation in an unsupervised manner. All of the segmentation algorithms are treated as new features and the segmentation results obtained by the algorithms are the values of the new features. The MMIBDF is designed to sample the boundary according to a discrete distribution. We present an inference method for MMIBDF and describe the corresponding algorithm in detail. Extensive empirical results demonstrate that MMIBDF significantly outperforms other image boundary detection fusion algorithms and the base image boundary detection algorithms according to most performance indices.

    Download PDF (800K)
  • Kento WATANABE, Yuichiroh MATSUBAYASHI, Kentaro INUI, Satoru FUKAYAMA, ...
    Type: PAPER
    Subject area: Natural Language Processing
    2018 Volume E101.D Issue 4 Pages 1167-1179
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper addresses the issue of modeling the discourse nature of lyrics and presented the first study aiming at capturing the two common discourse-related notions: storylines and themes. We assume that a storyline is a chain of transitions over topics of segments and a song has at least one entire theme. We then hypothesize that transitions over topics of lyric segments can be captured by a probabilistic topic model which incorporates a distribution over transitions of latent topics and that such a distribution of topic transitions is affected by the theme of lyrics. Aiming to test those hypotheses, this study conducts experiments on the word prediction and segment order prediction tasks exploiting a large-scale corpus of popular music lyrics for both English and Japanese (around 100 thousand songs). The findings we gained from these experiments can be summarized into two respects. First, the models with topic transitions significantly outperformed the model without topic transitions in word prediction. This result indicates that typical storylines included in our lyrics datasets were effectively captured as a probabilistic distribution of transitions over latent topics of segments. Second, the model incorporating a latent theme variable on top of topic transitions outperformed the models without such variables in both word prediction and segment order prediction. From this result, we can conclude that considering the notion of theme does contribute to the modeling of storylines of lyrics.

    Download PDF (1383K)
  • Yanxia QIN, Yue ZHANG, Min ZHANG, Dequan ZHENG
    Type: PAPER
    Subject area: Natural Language Processing
    2018 Volume E101.D Issue 4 Pages 1180-1188
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Large scale first-hand tweets motivate automatic event detection on Twitter. Previous approaches model events by clustering tweets, words or segments. On the other hand, event clusters represented by tweets are easier to understand than those represented by words/segments. However, compared to words/segments, tweets are sparser and therefore makes clustering less effective. This article proposes to represent events with triple structures called frames, which are as efficient as, yet can be easier to understand than words/segments. Frames are extracted based on shallow syntactic information of tweets with an unsupervised open information extraction method, which is introduced for domain-independent relation extraction in a single pass over web scale data. This is then followed by bursty frame element extraction functions as feature selection by filtering frame elements with bursty frequency pattern via a probabilistic model. After being clustered and ranked, high-quality events are yielded and then reported by linking frame elements back to frames. Experimental results show that frame-based event detection leads to improved precision over a state-of-the-art baseline segment-based event detection method. Superior readability of frame-based events as compared with segment-based events is demonstrated in some example outputs.

    Download PDF (418K)
  • Yande XIANG, Jiahui LUO, Taotao ZHU, Sheng WANG, Xiaoyan XIANG, Jianyi ...
    Type: PAPER
    Subject area: Biological Engineering
    2018 Volume E101.D Issue 4 Pages 1189-1198
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Arrhythmia classification based on electrocardiogram (ECG) is crucial in automatic cardiovascular disease diagnosis. The classification methods used in the current practice largely depend on hand-crafted manual features. However, extracting hand-crafted manual features may introduce significant computational complexity, especially in the transform domains. In this study, an accurate method for patient-specific ECG beat classification is proposed, which adopts morphological features and timing information. As to the morphological features of heartbeat, an attention-based two-level 1-D CNN is incorporated in the proposed method to extract different grained features automatically by focusing on various parts of a heartbeat. As to the timing information, the difference between previous and post RR intervels is computed as a dynamic feature. Both the extracted morphological features and the interval difference are used by multi-layer perceptron (MLP) for classifing ECG signals. In addition, to reduce memory storage of ECG data and denoise to some extent, an adaptive heartbeat normalization technique is adopted which includes amplitude unification, resolution modification, and signal difference. Based on the MIT-BIH arrhythmia database, the proposed classification method achieved sensitivity Sen=93.4% and positive predictivity Ppr=94.9% in ventricular ectopic beat (VEB) detection, sensitivity Sen=86.3% and positive predictivity Ppr=80.0% in supraventricular ectopic beat (SVEB) detection, and overall accuracy OA=97.8% under 6-bit ECG signal resolution. Compared with the state-of-the-art automatic ECG classification methods, these results show that the proposed method acquires comparable accuracy of heartbeat classification though ECG signals are represented by lower resolution.

    Download PDF (946K)
  • Jing LIU, Pei Dai XIE, Meng Zhu LIU, Yong Jun WANG
    Type: LETTER
    Subject area: Information Network
    2018 Volume E101.D Issue 4 Pages 1199-1202
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Malware phylogeny refers to inferring evolutionary relationships between instances of families. It has gained a lot of attention over the past several years, due to its efficiency in accelerating reverse engineering of new variants within families. Previous researches mainly focused on tree-based models. However, those approaches merely demonstrate lineage of families using dendrograms or directed trees with rough evolution information. In this paper, we propose a novel malware phylogeny construction method taking advantage of persistent phylogeny tree model, whose nodes correspond to input instances and edges represent the gain or lost of functional characters. It can not only depict directed ancestor-descendant relationships between malware instances, but also show concrete function inheritance and variation between ancestor and descendant, which is significant in variants defense. We evaluate our algorithm on three malware families and one benign family whose ground truth are known, and compare with competing algorithms. Experiments demonstrate that our method achieves a higher mean accuracy of 61.4%.

    Download PDF (686K)
  • Lianqiang LI, Yuhui XU, Jie ZHU
    Type: LETTER
    Subject area: Artificial Intelligence, Data Mining
    2018 Volume E101.D Issue 4 Pages 1203-1206
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper introduces a filter level pruning method based on similar feature extraction for compressing and accelerating the convolutional neural networks by k-means++ algorithm. In contrast to other pruning methods, the proposed method would analyze the similarities in recognizing features among filters rather than evaluate the importance of filters to prune the redundant ones. This strategy would be more reasonable and effective. Furthermore, our method does not result in unstructured network. As a result, it needs not extra sparse representation and could be efficiently supported by any off-the-shelf deep learning libraries. Experimental results show that our filter pruning method could reduce the number of parameters and the amount of computational costs in Lenet-5 by a factor of 17.9× with only 0.3% accuracy loss.

    Download PDF (182K)
  • Deokgyu YUN, Hannah LEE, Seung Ho CHOI
    Type: LETTER
    Subject area: Speech and Hearing
    2018 Volume E101.D Issue 4 Pages 1207-1208
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    This paper proposes a deep learning-based non-intrusive objective speech intelligibility estimation method based on recurrent neural network (RNN) with long short-term memory (LSTM) structure. Conventional non-intrusive estimation methods such as standard P.563 have poor estimation performance and lack of consistency, especially, in various noise and reverberation environments. The proposed method trains the LSTM RNN model parameters by utilizing the STOI that is the standard intrusive intelligibility estimation method with reference speech signal. The input and output of the LSTM RNN are the MFCC vector and the frame-wise STOI value, respectively. Experimental results show that the proposed objective intelligibility estimation method outperforms the conventional standard P.563 in various noisy and reverberant environments.

    Download PDF (312K)
  • Ying TIAN, Mingyong ZENG, Aihong LU, Bin GAO, Zhangkai LUO
    Type: LETTER
    Subject area: Image Recognition, Computer Vision
    2018 Volume E101.D Issue 4 Pages 1209-1212
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    A novel and efficient coding method is proposed to improve person re-identification in the XQDA subspace. Traditional CRC (Collaborative Representation based Classification) conducts independent dictionary coding for each image and can not guarantee improved results over conventional euclidian distance. In this letter, however, a specific model is separately constructed for each probe image and each gallery image, i.e. in probe-galley pairwise manner. The proposed pairwise-specific CRC method can excavate extra discriminative information by enforcing a similarity item to pull similar sample-pairs closer. The approach has been evaluated against current methods on two benchmark datasets, achieving considerable improvement and outstanding performance.

    Download PDF (90K)
  • Wonjun HWANG
    Type: LETTER
    Subject area: Image Recognition, Computer Vision
    2018 Volume E101.D Issue 4 Pages 1213-1216
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    In this letter, we propose a sequential convolutional residual network, where we first analyze a tangled network architecture using simplified equations and determine the critical point to untangle the complex network architecture. Although the residual network shows good performance, the learning efficiency is not better than expected at deeper layers because the network is excessively intertwined. To solve this problem, we propose a network in which the information is transmitted sequentially. In this network architecture, the neighboring layer output adds the input of the current layer and iteratively passes its result to the next sequential layer. Thus, the proposed network can improve the learning efficiency and performance by successfully mitigating the complexity in deep networks. We show that the proposed network performs well on the Cifar-10 and Cifar-100 datasets. In particular, we prove that the proposed method is superior to the baseline method as the depth increases.

    Download PDF (311K)
  • Jingwei YAN, Wenming ZHENG, Zhen CUI, Peng SONG
    Type: LETTER
    Subject area: Biocybernetics, Neurocomputing
    2018 Volume E101.D Issue 4 Pages 1217-1220
    Published: April 01, 2018
    Released: April 01, 2018
    JOURNALS FREE ACCESS

    Facial expressions are generated by the actions of the facial muscles located at different facial regions. The spatial dependencies of different spatial facial regions are worth exploring and can improve the performance of facial expression recognition. In this letter we propose a joint convolutional bidirectional long short-term memory (JCBLSTM) framework to model the discriminative facial textures and spatial relations between different regions jointly. We treat each row or column of feature maps output from CNN as individual ordered sequence and employ LSTM to model the spatial dependencies within it. Moreover, a shortcut connection for convolutional feature maps is introduced for joint feature representation. We conduct experiments on two databases to evaluate the proposed JCBLSTM method. The experimental results demonstrate that the JCBLSTM method achieves state-of-the-art performance on Multi-PIE and very competitive result on FER-2013.

    Download PDF (420K)
feedback
Top