Journal of Information Processing
Online ISSN : 1882-6652
Current issue
Showing 1-50 articles out of 65 articles from the selected issue
  • Sadao Obana
    Type: Special Issue of Intelligent Transportation Systems and Mobile Communication toward Super Smart Society
    2018 Volume 26 Pages 1
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (37K)
  • Katsuya Matsumoto, Susumu Ishihara
    Type: Special Issue of Intelligent Transportation Systems and Mobile Communication toward Super Smart Society
    Subject area: Network technology
    2018 Volume 26 Pages 2-10
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    Drivers can make driving plans and save traveling time if they can download real-time pictures taken at each driver's point of interest (POI). We designed a system that provides pictures taken by other drivers at a driver's POI. In one naive system design, vehicles send picture data to a central server via a cellular network whenever they take a picture. However, sending all pictures taken by vehicles to the server leads to high processing loads on the server and major amounts of cellular network traffic. In this paper, we propose a scalable data sharing system that uses VANETs and a cellular network for providing a picture taken at POI. For realizing the system, we propose a data upload control method. In this method, all vehicles exchange metadata via VANETs to know the existence of pictures taken by other vehicles and reduce the traffic for uploading pictures taken under a similar condition of other pictures. We evaluate the performance of the proposed method through simulations. The simulation results show that the proposed method reduces the cellular network traffic to 36% in comparison with a system that sends all pictures received by the server to meet all driver POI picture requests.

    Download PDF (1085K)
  • Le Tien Trien, Koichi Adachi, Yasushi Yamao
    Type: Special Issue of Intelligent Transportation Systems and Mobile Communication toward Super Smart Society
    Subject area: Network technology
    2018 Volume 26 Pages 11-19
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) based relay-assisted V2V scheme is effective in compensating shadowing loss and hence improves performance of ITS V2V communications, especially for non-line-of-sight (NLOS) areas such as intersections. However, when the number of vehicle stations becomes large, packet congestion happens frequently at the relay station (RS). It causes interruption of packet relaying transmission and limits the improvement by the scheme. This paper proposes a packet payload combining relay (PCRL) scheme to alleviate the congestion issue at RS by compressing packet overheads of broadcast packets. With adaptive modulation and coding technique at RS, PCRL can further mitigate the packet interruption by reducing the air time of relayed packets. Time division grouping method is also applied to PCRL to mitigate the hidden terminal problem at intersections. Analyzed and simulated results show that the proposed schemes can effectively mitigate packet congestion issue and improve the reliability of the V2V communication.

    Download PDF (1232K)
  • Nurul Ain Binti Adnan, Shigeru Yamashita
    Type: Regular Papers
    Subject area: Innovative Computing Methods
    2018 Volume 26 Pages 20-28
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    This paper formulates the qubit layout problem for ICM representation which is favorable for topological quantum computation. Observing the properties of braiding operations in a logic circuit model of ICM, we study the potential usefulness of two-dimensional qubit layouts based on this model. We compare and contrast the efficiency of one- and two-dimensional qubit layouts in reducing the logical time steps for topological quantum circuits. This leads us to an approach to find a good gate order for two-dimensional qubit layouts. Indeed, our preliminary experimental results show the effectiveness of two-dimensional qubit layouts.

    Download PDF (3345K)
  • Takahiko Kato, Masaki Bandai
    Type: Regular Papers
    Subject area: Network Architecture
    2018 Volume 26 Pages 29-37
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    In the paper, we propose a new congestion control method based on hop-by-hop rate control in named data networking (NDN). The proposed method suppresses the excessive reduction of the Interest sending rate due to continuous negative acknowledgement (NACK) packets caused by the propagation delay. The proposed method limits the rate reduction to once per congestion. The point of the proposed method is to keep the Interest sending rate for NACK packets unchanged after the rate reduction for congestion until the rate reduction becomes effective. Performance evaluation is conducted through computer simulations in order to demonstrate that the proposed method improves the throughput performance as compared to the conventional method.

    Download PDF (1807K)
  • Hiroshi Yamada, Kazuya Murao, Tsutomu Terada, Masahiko Tsukamoto
    Type: Regular Papers
    Subject area: Mobile Computing
    2018 Volume 26 Pages 38-47
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    Competitive karuta is an official Japanese card game and is described as “martial art on the tatami.” Recently, competitive karuta has attracted a great deal of attention among young people. One of characteristic rules of competitive karuta is that there is no referee; therefore players must judge themselves even if the difficult situation arises. Consequently, the players sometimes get into an argument over their judgement, which disrupts the other matches in the room because all the matches proceed in parallel. In this paper, we propose a system that judges the player who took a card first in a competitive karuta match. Our system measures motion data when players take a card by using a wrist-worn accelerometer and gyroscope, and estimates the times when the players touched the card. From the evaluation experiments, 69.2% of rounds were estimated without error and 99.0% of rounds were estimated within 20-ms error. When our system was introduced on the close game, the accuracy of deciding the player taking a card was 75%.

    Download PDF (1633K)
  • Izumi Asakura, Hidehiko Masuhara, Takuya Matsumoto, Kiminori Matsuzaki
    Type: Regular Papers
    Subject area: Special Section on Programming
    2018 Volume 26 Pages 48-53
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    This paper supplements the evaluation of libraries for parallel computing in Haskell by Matsumoto and Matsuzaki. The previous work implemented an image processing program by using two different libraries and compared expressiveness and parallel execution performance. In particular, it found that a parallel execution of the Accelerate program on a GPU is slower than the execution of the handwritten CUDA-C program by a factor of more than 70. We analyzed the execution of the Accelerate program and identified that the major sources of the overhead are in the process of the syntax tree before execution of a GPU kernel. Since the program realizes repeated execution by constructing a large syntax tree whose size is proportional to the number of repetitions, we rewrote this part by using a while-construct, which significantly reduces the overhead as it runs half as fast as the CUDA-C program. We also discuss language design alternatives for avoiding this type of overhead.

    Download PDF (574K)
  • Akira Tanaka, Reynald Affeldt, Jacques Garrigue
    Type: Regular Papers
    Subject area: Special Section on Programming
    2018 Volume 26 Pages 54-72
    Published: 2018
    Released: January 15, 2018
    JOURNALS FREE ACCESS

    Our goal is the production of formally-verified pieces of low-level code. Low-level code is typically written in C, so as to enable efficient manipulation of data at the bit-level and easy access to built-in features of CPUs. Proof-assistants arguably provide the most rigorous approach to formal verification of computer programs. Unfortunately, they only allow for extraction of runnable code in high-level languages such as ML. Of course it is possible to embed C snippets into ML programs, but this results in a complicated extraction process and the performance of the output program becomes difficult to anticipate. In this paper, we propose a new code generation scheme for the Coq proof-assistant that directly generates provably-safe C code. It is implemented in the form of plugins. The generation of C source code is done by a plugin performing beforehand monomorphization of Coq programs. The correctness of monomorphization can be proved within Coq. Code generation allows for user-guided changes of data structures. It is therefore possible to do formal verification using proof-friendly data structures, while enjoying optimized C representations in the output code. In order to ensure the safety of this transformation, we propose a new customizable monadification algorithm in the form of another plugin. Using monadification, one can ensure by the insertion of the right monads the preservation of critical invariants, such as the absence of overflows or complexity properties. We provide several examples to illustrate our approach, including a realistic use-case: the rank algorithm from succinct data structures.

    Download PDF (780K)
  • Tatsuki Sekino
    Type: Special Issue of Computer and Humanities
    2018 Volume 26 Pages 73
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (40K)
  • Takeshi Miura, Takaaki Kaiga, Takeshi Shibata, Madoka Uemura, Katsubum ...
    Type: Special Issue of Computer and Humanities
    Subject area: Body Information Processing
    2018 Volume 26 Pages 74-84
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    This study proposes a method to systematically visualize the motion-characteristic distribution of Japanese folk dances passed down in a certain area. This is accomplished by adopting an approach that involves analyzing motion-capture data collected from the dances. The visualization process in the proposed method consists of three stages. The first stage is the modeling of the relationship among motion-capture data, folk dances, and the settlements in which folk dances have been passed down. This relationship is modeled as a hierarchical-structure model. The second stage is the extraction of motion characteristics from motion-capture data streams. The motion characteristics of each data stream are summarized as a fourteen-dimensional feature vector. The third stage is the visualization of the motion-characteristic distribution of the dances investigated. Each of the dances is mapped on a two-dimensional scatter plot in accordance with the feature quantities obtained in the second stage. Information on the hierarchical-structure model constructed in the first stage is also displayed. The analysis results for the distribution of Bon Odori dances showed that the proposed method could have almost completely visualized the motion-characteristic distribution of sample folk dances, while also demonstrating consistency with the knowledge of the dances acquired in the previous studies.

    Download PDF (2532K)
  • Yuuichi Teranishi
    Type: Special Issue of Network and Distributed Processing
    2018 Volume 26 Pages 85-86
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (38K)
  • Hiroyuki Tanaka, Masahiro Yoshida, Koya Mori, Noriyuki Takahashi
    Type: Special Issue of Network and Distributed Processing
    Subject area: Invited Papers
    2018 Volume 26 Pages 87-97
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Multi-access Edge Computing (MEC) can be defined as a model for enabling business oriented, cloud computing platform within multiple types of the access network (e.g., LTE, 5G, WiFi, FTTH, etc.) at the close proximity of subscribers to serve delay sensitive, context aware applications. To pull out the most of the potential, MEC has to be designed as infrastructure, to support many kind of IoT applications and their eco system, in addition to sufficiently management mechanism. In this context, various research and standardization efforts are ongoing. This paper provides a comprehensive survey of the state-of-the-art research efforts on MEC domain, with focus on the architectural proposals as infrastracture, the issue of the partitioning of processing among the user devices, edge servers, and a cloud, and the issue of the resource management.

    Download PDF (950K)
  • Babatunde Ojetunde, Naoki Shibata, Juntao Gao
    Type: Special Issue of Network and Distributed Processing
    Subject area: Implementation of Operating System Functionality
    2018 Volume 26 Pages 98-110
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Secure communications is essential in many areas such as disaster management and battlefield communications. To detect and prevent attacks in such applications, most existing protocols adopt a cryptography-based approach, trust-based approach (reputation of nodes), or incentive-based approach. However, such protocols still have drawbacks, such as expensive overhead, difficulty in maintaining secure key and session management, or leaving routes unsecured against Byzantine attacks. In this paper, we introduce a monitoring-based method in the link state routing protocol to secure the packets' route against Byzantine attacks. The goal of our proposed scheme is to guarantee communication among connected benign nodes in the network. Specifically, each node monitors the action of neighboring nodes and compares the optimal packet route against the packet route history. Nodes in the network create a packet history field which is used to record all activities of an intermediate node when receiving and forwarding packets. Our scheme provides mutual monitoring in which nodes in the network can validate the packet history field of other nodes and report malicious activities. Also, our scheme uses a statistical method to know if a node is dropping packets intentionally by analyzing the packet dropping behavior of each node. The proposed scheme provides protection against colluding attacks and other Byzantine attacks. The proposed monitoring-based method achieves an average of 89% to 96% packet delivery ratio when 11% to 21% active malicious links are excluded from the network.

    Download PDF (2258K)
  • Arif Herusetyo Wicaksono, Kazuhide Aikoh
    Type: Special Issue of Network and Distributed Processing
    Subject area: Data Models and Database Design
    2018 Volume 26 Pages 111-115
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    With the needs of consistency between multiple datacenters for state management in transaction processing system, multi-datacenter synchronization is required, both for reducing response time and for providing fail-over in the event of disaster. However, in synchronization, prioritizing consistency causes latency degradation, while preserving response times may cause data inconsistencies, both of which are unacceptable. Therefore, we propose a synchronization method with two characteristics: prioritization of identifier transmission along with transaction identifier, and replication status verification prior to access. By applying the proposal to prototype in-memory KVS, we verified the response time degradation can be reduced while eliminating potential of conflicts.

    Download PDF (485K)
  • Kiyotaka Kaji, Takuya Yoshihiro
    Type: Special Issue of Network and Distributed Processing
    Subject area: Network Protocols
    2018 Volume 26 Pages 116-123
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    MANETs (Mobile Ad-hoc NETworks) have been studied for a long time to provide multi-hop wireless communications among temporarily gathered mobile devices. However, due to the nature of wireless communications, the communication performance of a flow is easily degraded by even a single local congestion on the path. A simple solution for this problem is to utilize detour paths that avoid the congested area. However, to provide detour paths that avoid the area whenever a packet faces congestion is in fact a challenging task because the detour paths must avoid relatively large region in which harmful radio waves interfere. In this paper, we propose a routing scheme that proactively computes detour paths to avoid the congested area ahead of packets. Specifically, when a packet meets congestion, we assume that the next-next-hop node along the shortest path to the destination is the center of the congestion, and forward the packet to its destination without visiting the node neighboring the congestion central node. We designed an algorithm that builds the secondary routing tables that works to provide the detour paths on top of the distributed link-state routing scheme.

    Download PDF (719K)
  • Yuria Oigawa, Fumiaki Sato
    Type: Special Issue of Network and Distributed Processing
    Subject area: Network Protocols
    2018 Volume 26 Pages 124-131
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    A mobile ad hoc Network (MANET) is a self-organizing and self-configurable wireless network of mobile nodes, which move arbitrarily without a pre-existing infrastructure. Routing is a critical issue in MANETs. One well-known protocol for ad hoc networks is the Zone Routing Protocol (ZRP). However, because ZRP sends many useless control packets, the network load increases and the network performance decreases. This paper examines a novel routing protocol that uses Bloom filters and the topology information of neighbor nodes to improve ZRP. A Bloom filter is often used as a content discovery method. In this case, the content ID is hashed to one value. In our proposed protocol, the node ID is hashed to one value, and a Bloom filter is generated from the ID values in a zone. The generated Bloom filter is propagated across a tree structure over a MANET. The tree is constructed by the peripheral nodes of the IntErzone Routing Protocol (IERP) of ZRP. The Bloom Filter sent from a peripheral node is managed as node information of the direction of the peripheral node. The number of RouteQuery packets of IERP can be reduced by using the managed Bloom filter. We compare the performance of the proposed routing protocol with the bordercast of IERP and show the feasibility of the proposed method as a routing protocol for MANETs.

    Download PDF (1363K)
  • Ushio Yamamoto
    Type: Special Issue of Network and Distributed Processing
    Subject area: Wireless/Mobile Networks
    2018 Volume 26 Pages 132-139
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Geographic routing is one of the most effective routing methods for ad-hoc wireless networks, and each node selects the forwarding node to the destination node by using position information of its neighbor nodes. However, it has a problem that forwarded packets may encounter the void area where no closer node to the destination is. In this paper, we propose a geographic routing method with landmarks between source and destination node, called GLGR (Grid Landmark-based Geographic Routing). To handle node mobility, our proposed method employs grid-based landmarks, not node-based landmarks used in the traditional methods. A grid is a divided region of communication area and its position is static. Therefore, landmarks can keep the appropriate position to forward data packets to the destination, especially in the case that the void is static. Also, to adaptively change the cache interval of landmark information, the node guesses whether the encountered void is static or dynamic by using the node habitability of grids. Network simulation results about packet arrival rate and end-to-end delay showed 30% better performance of our proposed method than the traditional methods in the best case, with suppressing control overhead.

    Download PDF (735K)
  • Akihiro Fujimoto, Yukari Masui, Takuya Yoshihiro
    Type: Special Issue of Network and Distributed Processing
    Subject area: Wireless/Mobile Networks
    2018 Volume 26 Pages 140-147
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Many low-energy MAC and routing protocols have been proposed for Wireless Sensor Networks (WSNs) since reducing energy consumption is a primary concern to meet the requirements of practical applications. Reducing delivery delay is, indeed, another promising requirement in WSNs because a large part of applications of WSNs intends to watch the sensed objects in real time although the amount of allowable delay differs according to cases. In this paper, we deal with a class of receiver-initiated MAC protocols, which is a promising energy-efficient MAC mechanism, and propose to schedule beacon timings so as to reduce the delivery delay of data packets. The key idea is to schedule beacon timings in the sequence of distance from leaf node to the sink node and reduce the time of packets staying at each node. Specifically, each sensor node selects a time slot to send beacon in a distributed manner based on their distances (i.e., hop counts) from the sink node, and autonomously adjusts its beacon timing to avoid frame collisions. Computer simulations show that the proposed protocol can collect data in a shorter time with less energy consumption than the conventional receiver-initiated MAC protocols.

    Download PDF (885K)
  • Jun Liu, Kensuke Fukuda
    Type: Special Issue of Network and Distributed Processing
    Subject area: Network Quality and Control
    2018 Volume 26 Pages 148-157
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    To enhance Internet security, researchers have largely emphasized diverse cyberspace monitoring approaches to observe cyber attacks and anomalies. Among them darknet provides an effective passive monitoring one. Darknets refer to the globally routable but still unused IP address spaces. They are often used to monitor unexpected incoming network traffic, and serve as an effective network traffic measurement approach for viewing certain remote network security activities. Previous works in this field discussed possible causes (i.e., anomalies) of darknet traffic and applied their classification schemes on short-term traces. Our interest lies, however, in how darknet traffic has evolved and the effectiveness of a darknet traffic taxonomy for longitudinal data. To reach these goals, we propose a simple darknet traffic taxonomy based on network traffic rules, and evaluate it with two darknet traces: one covering 12 years since 2006, while the other covering 11 years since 2007. The evaluation results reveal the effectiveness of this taxonomy: we are able to label over 94% of all source IPs with anomalies defined by the taxonomy, leaving the unlabeled source ratio low. We also examine the evolution of different anomalies since 2006 (especially in recent years), analyze the temporal and spatial dependency and parameter dependency of darknet traffic, and conclude that most sources in the datasets are characterized by just one or two anamalies with simple attack mechanisms. Moreover, we compare the taxonomy with a one-way traffic analysis tool (i.e., iatmon) to better understand their differences.

    Download PDF (593K)
  • Ei Khaing Win, Tomoki Yoshihisa, Yoshimasa Ishi, Tomoya Kawakami, Yuui ...
    Type: Special Issue of Network and Distributed Processing
    Subject area: Network Services
    2018 Volume 26 Pages 158-168
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    In this paper, we propose a lost data recovery scheme called the synchronized recovery stream merging (SRSM) for sensor data stream multicasting in the unstable networks of the IoT applications. We propose two types of SRSM. The first one is “Latency-aware synchronized recovery stream merging (SRSM-L)” for IoT applications that has restrictions of acceptable latency. The other is “Bandwidth-dependent synchronized recovery stream merging (SRSM-B)” for the cases in which the network bandwidth for the sender is limited. Our proposed schemes reduce the number of the streams managed by the sender by waiting for other recovery streams to synchronize the delivery timing and merging. From our simulation evaluations, we confirmed that our proposed schemes save network bandwidth on the sender in the random and burst failure situations. We confirmed that SRSM-L could reduce the network bandwidth of the sender about 52% in the random failure situation, keeping the acceptable latency. We also confirmed that SRSM-B could keep the specified number of streams constant and the latency overheads small in the burst failure situation.

    Download PDF (1336K)
  • Tomoo Inoue
    Type: Advances in Collaboration Technologies
    2018 Volume 26 Pages 169
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (35K)
  • Stephen J.H. Yang, Owen H.T. Lu, Anna Y.Q. Huang, Jeff C.H. Huang, Hir ...
    Type: Advances in Collaboration Technologies
    Subject area: Invited Papers
    2018 Volume 26 Pages 170-176
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    With the rise of big data analytics, learning analytics has become a major trend for improving the quality of education. Learning analytics is a methodology for helping students to succeed in the classroom; the principle is to predict student's academic performance at an early stage and thus provide them with timely assistance. Accordingly, this study used multiple linear regression (MLR), a popular method of predicting students' academic performance, to establish a prediction model. Moreover, we combined MLR with principal component analysis (PCA) to improve the predictive accuracy of the model. Traditional MLR has certain drawbacks; specifically, the coefficient of determination (R2) and mean square error (MSE) measures and the quantile-quantile plot (Q-Q plot) technique cannot evaluate the predictive performance and accuracy of MLR. Therefore, we propose predictive MSE (pMSE) and predictive mean absolute percentage correction (pMAPC) measures for determining the predictive performance and accuracy of the regression model, respectively. Analysis results revealed that the proposed model for predicting students' academic performance could obtain optimal pMSE and pMAPC values by using six components obtained from PCA.

    Download PDF (1339K)
  • Bektur Ryskeldiev, Michael Cohen, Jens Herder
    Type: Advances in Collaboration Technologies
    Subject area: Group Interaction Support and Groupware
    2018 Volume 26 Pages 177-185
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    We present a system that exploits mobile rotational tracking and photospherical imagery to allow users to share their environment with remotely connected peers “on the go.” We surveyed related interfaces and developed a unique groupware application that shares a mixed reality space with spatially-oriented live video feeds. Users can collaborate through realtime audio, video, and drawings in a virtual space. The developed system was tested in a preliminary user study, which confirmed an increase in spatial and situational awareness among viewers as well as reduction in cognitive workload. Believing that our system provides a novel style of collaboration in mixed reality environments, we discuss future applications and extensions of our prototype.

    Download PDF (4906K)
  • Jesús Omar Álvarez Márquez, Jürgen Ziegler
    Type: Advances in Collaboration Technologies
    Subject area: Group Interaction Support and Groupware
    2018 Volume 26 Pages 186-200
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    This article presents an approach to group recommender systems that focuses its attention on the group's social interaction during the formulation, discussion and negotiation of the features the item to be jointly selected should possess. Current group recommender techniques are mainly based on aggregating existing user profiles or on a profile of the group as a whole. Our method supports collaborative preference elicitation and negotiation process where desired features have to be chosen individually, but group consensus is needed for them to become active in the item filtering process. Users provide feedback on the selected preferences and change their significance, bringing up new recommendations each time individual settings are modified. The last stage in the decision process is also supported, when users collectively select the final item from the recommendation set. We explored the possible benefits of our approach through the development of three prototypes, each based on a different variant of the approach with a different emphasis on private and group-wide preference spaces. They were evaluated with user groups of different size, addressing questions regarding the effectiveness of different information sharing methods and the repercussion of group size in the recommendation process. We compare the different methods and consolidate the findings in an initial model of recommending for group.

    Download PDF (998K)
  • Eiichi Hayakawa
    Type: Special Issue of Embedded Systems Engineering
    2018 Volume 26 Pages 201
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (32K)
  • Hiroyuki Chishiro
    Type: Special Issue of Embedded Systems Engineering
    Subject area: Embedded System Technology
    2018 Volume 26 Pages 202-211
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Optimal multiprocessor real-time scheduling can achieve full system utilization with implicit-deadline periodic task sets. However, worst case execution time (WCET) analysis is difficult on state-of-the-art hardware/software platforms due to the complex hierarchy of shared caches and multiprogramming. The actual case execution time of each task is usually shorter than its WCET and imprecise computation is an effective method to make better use of the remaining processor time. Semi-fixed-priority scheduling is real-time scheduling that supports imprecise computation and multiprocessors but conventional semi-fixed-priority scheduling algorithms are not optimal. This paper proposes an optimal multiprocessor semi-fixed-priority scheduling algorithm that supports imprecise computation. The proposed algorithm, which integrates Reduction to Uniprocessor (RUN) for Rate Monotonic with Wind-up Part (RMWP), called RUN-RMWP, is superior to Partitioned RMWP algorithm in terms of schedulability analysis. Simulation studies show that RUN-RMWP has a few more preemptions/migrations compared to RUN but confirms its optimality even though conventional semi-fixed priority scheduling algorithms are not optimal.

    Download PDF (1053K)
  • Bo Sun, Xiapu Luo, Mitsuaki Akiyama, Takuya Watanabe, Tatsuya Mori
    Type: Regular Papers
    Subject area: System Security
    2018 Volume 26 Pages 212-223
    Published: 2018
    Released: February 15, 2018
    JOURNALS FREE ACCESS

    Mobile app stores, such as Google Play, play a vital role in the ecosystem of mobile device software distribution platforms. When users find an app of interest, they can acquire useful data from the app store to inform their decision regarding whether to install the app. This data includes ratings, reviews, number of installs, and the category of the app. The ratings and reviews are the user-generated content (UGC) that affect the reputation of an app. Therefore, miscreants can leverage such channels to conduct promotional attacks; for example, a miscreant may promote a malicious app by endowing it with a good reputation via fake ratings and reviews to encourage would-be victims to install the app. In this study, we have developed a system called PADetective that detects miscreants who are likely to be conducting promotional attacks. Using a 1723-entry labeled dataset, we demonstrate that the true positive rate of detection model is 90%, with a false positive rate of 5.8%. We then applied our system to an unlabeled dataset of 57M reviews written by 20M users for 1M apps to characterize the prevalence of threats in the wild. The PADetective system detected 289K reviewers as potential PA attackers. The detected potential PA attackers posted reviews to 136K apps, which included 21K malicious apps. We also report that our system can be used to identify potentially malicious apps that have not been detected by anti-virus checkers.

    Download PDF (972K)
  • Hayato Yamana
    Type: Special Issue of Young Researchers' Papers
    2018 Volume 26 Pages 224
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (35K)
  • Yixiao Li, Yutaka Matsubara, Hiroaki Takada
    Type: Special Issue of Young Researchers' Papers
    Subject area: Embedded System Technology
    2018 Volume 26 Pages 225-236
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Current embedded systems are usually based on real-time operating system (RTOS). In the near future, embedded systems will include parallel applications for tasks like autonomous driving, and adopt many-core processors to satisfy the performance requirements. However, traditional RTOSes are not designed for high performance applications and whether they can scale well on many-core processors remains unclear. Meanwhile, research has shown that Linux can provide good scalability for processors with tens of cores. In this paper, an experiment environment based on a traditional multi-core RTOS (TOPPERS/FMP) and an off-the-shelf 72-core many-core processor (TILE-Gx72) is presented. By a comparative analysis of RTOS based and Linux based runtime systems, several bottlenecks in RTOS are identified and the methods to avoid them are proposed. After that, the PARSEC benchmark suite is used to evaluate the performance of RTOS and Linux. The results show that the optimized RTOS runtime system tends to deliver better scalability than Linux in many cases. Therefore, we believe that traditional RTOS like TOPPERS/FMP can still be a good choice for embedded many-core processors in the near future.

    Download PDF (2135K)
  • Jiachao Zhang, Shinji Fukuma, Shin-ichiro Mori
    Type: Special Issue of Young Researchers' Papers
    Subject area: Global Distributed Computing Systems
    2018 Volume 26 Pages 237-246
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Diverse simulations on centralized high performance computers are the driving force behind innovation in nearly all fields. Higher resolution and real-time interactions of users are expected to reveal valuable phenomena by providing more precise details and efficient dynamic steering inside simulations. If simulations are performed at a centralized server, such as supercomputer or cloud system, it is difficult to satisfy the requirements of both real-time interaction and high-resolution for many of those user's domain of interest (DoI). To solve this problem, we utilize distributed regional servers closer to users to perform simulations for each independent DoI region. However, this may introduce inaccuracies in the boundary condition for each DoI region. In order to improve the boundary conditions for better regional simulations, the servers must cooperate to exchange necessary information. For this reason, this paper proposes a general purpose framework named Silk Road, to help application users realize a distributed collaborative simulation which utilizes regional servers to perform high-resolution simulation with low network delay, with help from exchanging boundary conditions through collaborations with each other via a central server which performs a low-resolution but wider area simulation to couple the regional simulations to propagate the higher resolution simulation results. The most notable feature of Silk Road Framework is the bi-directional refinement of on-going simulation by exchanging an adequate amount of simulation results occasionally without inducing unacceptable network delays. Through a case study, with a 2D diffusion simulation, we show the framework can achieve the distributed collaborative simulation and our model can help refine simulations on both the regional and central server side.

    Download PDF (4477K)
  • Isakwisa Gaddy Tende, Shin-Ichiro Kubota, Hisaaki Yamaba, Kentaro Abur ...
    Type: Special Issue of Young Researchers' Papers
    Subject area: Network Services
    2018 Volume 26 Pages 247-256
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    A large part of the population in Tanzania consists of rural smallholder farmers. Due to lack of market information, farmers sell crops to middlemen at lower prices and hence don't get enough profits. Previously we developed a WEB and SMS (Short Message Service) based Farmers Market Information System using a Waterfall model which automatically replies to farmers' requests for market prices of crops through SMS. We then improved the system so that, farmers can access crop buyers' details and crop demand reports through SMS while crop buyers can access farmers' details and crop demand reports through SMS or WEB. We also evaluated the effectiveness of the system to crop buyers' incomes and times for accessing market information after implementing the system in Tanzania in the SWAHILI language. In this paper we further evaluate the system effectiveness. Findings reveal that the system helped farmers to get improved incomes and access market information in shorter times. The system also helped farmers and crop buyers to sell and buy larger quantities of crops respectively. The findings also reveal that young people can learn and use the system more easily than old people. Users can also easily learn and use the system if they can read and write in the SWAHILI language regardless of their education levels.

    Download PDF (1546K)
  • Shotaro Usuzaki, Yuki Arikawa, Hisaaki Yamaba, Kentaro Aburada, Shin-I ...
    Type: Special Issue of Young Researchers' Papers
    Subject area: Network Security
    2018 Volume 26 Pages 257-266
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Distributed Denial-of-Service (DDoS) attack detection systems are classified into a signature based approach and an anomaly based approach. However, such methods tend to suffer from low responsiveness. On the other hand, real-time burst detection which is used in data mining offers two advantages over traditional statistical methods. First, it can be used for real-time detection when an event is occurring, and second, it can work with less processing as information about events are compressed, even if a large number of events occur. Here, the authors add the function for attack detection in real-time burst detection technique, and propose a highly responsive DDoS attack detection technique. This paper performs experiments to evaluate its effectiveness, and discusses its detection accuracy and processing performance.

    Download PDF (1026K)
  • Yo Ehara, Issei Sato, Hidekazu Oiwa, Hiroshi Nakagawa
    Type: Special Issue of Young Researchers' Papers
    Subject area: Natural Language Processing
    2018 Volume 26 Pages 267-275
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    While there have been many studies on measuring the size of learners' vocabulary or the vocabulary they should learn, there have been few studies on what kind of words learners think that they know. Therefore, we investigated theoretically and practically important models for predicting second language learners' vocabulary and propose another model for this vocabulary prediction task. With the current models, the same word difficulty measure is shared by all learners. This is unrealistic because some learners have special interests. A learner interested in music may know special music-related terms regardless of their difficulty. To solve this problem, our model can define a learner-specific word difficulty measure. Our model is also an extension of these current models in the sense that these models are special cases of our model. In a qualitative evaluation, we defined a measure for how learner-specific a word is. Interestingly, the word with the highest learner-specificity was “twitter.” Although “twitter” is a difficult English word, some low-ability learners presumably knew this word through the famous micro-blogging service. Our qualitative evaluation successfully extracted such interesting and suggestive examples. Our model achieved an accuracy competitive with the current models.

    Download PDF (312K)
  • Tatsunori Hirai, Hironori Doi, Shigeo Morishima
    Type: Special Issue of Young Researchers' Papers
    Subject area: Music Information Processing
    2018 Volume 26 Pages 276-284
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    This paper presents a topic modeling method to retrieve similar music fragments and its application, Music-Mixer, which is a computer-aided DJ system that supports DJ performance by automatically mixing songs in a seamless manner. MusicMixer mixes songs based on audio similarity calculated via beat analysis and latent topic analysis of the chromatic signal in the audio. The topic represents latent semantics on how chromatic sounds are generated. Given a list of songs, a DJ selects a song with beats and sounds similar to a specific point of the currently playing song to seamlessly transition between songs. By calculating similarities between all existing song sections that can be naturally mixed, MusicMixer retrieves the best mixing point from a myriad of possibilities and enables seamless song transitions. Although it is comparatively easy to calculate beat similarity from audio signals, considering the semantics of songs from the viewpoint of a human DJ has proven difficult. Therefore, we propose a method to represent audio signals to construct topic models that acquire latent semantics of audio. The results of a subjective experiment demonstrate the effectiveness of the proposed latent semantic analysis method. MusicMixer achieves automatic song mixing using the audio signal processing approach; thus, users can perform DJ mixing simply by selecting a song from a list of songs suggested by the system.

    Download PDF (2383K)
  • Marimo Kumaki, Yoshinari Takegawa, Keiji Hirata
    Type: Special Issue of Young Researchers' Papers
    Subject area: User Interfaces and Interactive Systems
    2018 Volume 26 Pages 285-293
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    In this paper, we propose a positioning learning support system with false information and vague information teaching functions for violin beginners. A hypothesis is made that the system obscures presented information from the learner or presents erroneous information so that the learner can break away from the learning support information. In an experiment using this system, it was shown that the group learning with vague information and false information tailored to the proficiency level learned more effectively than the group that continued learning with only accurate learning support information. This demonstrates the usefulness of the system. Moreover, we structured the causal relationship between True mode, Vague mode, and False mode objectively and visually in the proposed system using a structured modeling method using the ISM method (Interpretive Structural Modeling). A mutual relationship between various modes can be illustrated as with a multilayer directed graph. From these results, we consider the ideal learning model for the casual relationship between each mode.

    Download PDF (5247K)
  • Mari Ito, Masatake Hirao, Hiroki Hamahara
    Type: Special Issue of Young Researchers' Papers
    Subject area: Medicine and Welfare
    2018 Volume 26 Pages 294-300
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    This paper introduces a 0-1 integer programming model for scheduling nursery staff shifts. A support system is designed to manage challenges such as satisfying nursery school requests and reducing the scheduling workload. The scheduling goals are to reduce the number of times per month nursery staff are given the same shift as well as the numbers of consecutive early, late, and identical shifts. To achieve this, we place weighted penalties on the constraints requested by the nursery school and use an objective function that minimized these penalties. The schedule obtained by the support system improved the management of nursery staff shift scheduling, and comparisons with schedules drawn up by hand confirmed the superiority of those produced by the system.

    Download PDF (491K)
  • Jiachao Zhang, Shin-ichiro Mori, Shinji Tomita
    Type: Regular Papers
    Subject area: Parallel and Distributed Processing Technology
    2018 Volume 26 Pages 301-305
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    For medium scale commodity cluster, a high cost performance interconnection network named Three Quads was proposed. Topological features of Three Quads include 1) diameter of two, 2) embed-ability of 3D torus and many well-known topologies, and 3) fault tolerance and path diversity. All the hardware components of Three Quads are essentially small-scale and off-the-shelf commodity hardware to achieve superior cost performance. In this paper, the performance of the Three Quads is evaluated using a matrix transpose application and results show that the Three Quads gives appealing performance comparable to a large scale and expensive enterprise switch fabric.

    Download PDF (2057K)
  • Takuya Kuwahara, Yukino Baba, Hisashi Kashima, Takeshi Kishikawa, Juni ...
    Type: Regular Papers
    Subject area: Machine Learning & Data Mining
    2018 Volume 26 Pages 306-313
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Modern vehicles are equipped with Electronic Control Units (ECUs) and external communication devices. The Controller Area Network (CAN), a widely used communication protocol for ECUs, does not have a security mechanism to detect improper packets; if attackers exploit the vulnerability of an ECU and manage to inject a malicious message, they are able to control other ECUs to cause improper operation of the vehicle. With the increasing popularity of connected cars, it has become an urgent matter to protect in-vehicle networks against security threats. In this paper, we study the applicability of statistical anomaly detection methods for identifying malicious CAN messages in in-vehicle networks. We focus on intrusion attacks of malicious messages. Because the occurrence of an intrusion attack certainly influences the message traffic, we focus on the number of messages observed in a fixed time window to detect intrusion attacks. We formalize features to represent a message sequence that incorporates the number of messages associated with each receiver ID. We collected CAN message data from an actual vehicle and conducted a quantitative analysis of the methods and the features in practical situations. The results of our experiments demonstrated our proposed methods provide fast and accurate detection in various cases.

    Download PDF (725K)
  • Kosuke Matsumoto, Tomoharu Ugawa, Tatsuya Abe
    Type: Regular Papers
    Subject area: Special Section on Programming
    2018 Volume 26 Pages 314-326
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Modern multi-core CPUs might execute memory access instructions of programs out-of-order. However, the SPIN model checker does not check out-of-order executions: it only checks in-order executions. We have developed a library for SPIN that enables checking such out-of-order executions with respect to two memory models, the total store ordering (TSO) and the partial store ordering (PSO). This library provides models of variables shared with multiple threads (shared variables), and read and write macros to access them. Nevertheless, this library has three problems. First, although SPIN accepts Linear Temporal Logic (LTL) formulas, which are used for representing properties to be checked such as safety and liveness, our library did not support LTL formulas referring to shared variables. Secondly, guard statements, which are often used for blocking threads while a guard is not executable, cannot refer to shared variables. Finally, the user was unable to specify initial values of shared variables, but they are initialized with zero. As presented herein, we improved the library to resolve these problems. We produced models using our improved library and investigated the library performance.

    Download PDF (781K)
  • Wei Zhang, Shigeru Chiba
    Type: Regular Papers
    Subject area: Special Section on Programming
    2018 Volume 26 Pages 327-334
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    We propose call-trail dependent inline caching to improve the method dispatch performance of Method Seals. Method Seals is a class extension mechanism that allows users to manually control the effective range of class extensions. It provides better safety than existing class extension mechanisms. However, the absence of inline method cache renders Method Seals' runtime performance unsatisfactory. To enable inline caching on Method Seals, we added call-trail dependency to the conventional inline caching mechanism. To that end, we introduced the notion of call-trails which represent sets of classes along a call path. We use fixed-length bitsets for representing the current call-trail and a method definition's unsealed package list. Also, we relaxed Method Seals' semantic constraints accordingly in order to implement our proposal. We also implemented the proposed call-trail dependent inline caching on top of Method Seals and benchmarked its performance.

    Download PDF (679K)
  • Yuta Sugimoto, Atusi Maeda
    Type: Regular Papers
    Subject area: Special Section on Programming
    2018 Volume 26 Pages 335-344
    Published: 2018
    Released: March 15, 2018
    JOURNALS FREE ACCESS

    Packrat parsing is a recursive descent parsing method with backtracking and memoization. Parsers based on this method require no separate lexical analyzers, and backtracking enables those parsers to handle a wide range of complex syntactic constructs. Memoization is used to prevent exponential growth of running time, resulting in linear time complexity at th cost of linear space consumption. In this study, we propose CPEG - a library that can be used to write parsers using Packrat parsing in C language. This library enables programmers to describe syntactic rules in an internal domain-specific language (DSL) which, unlike parser combinators, does not require runtime data structures to represent syntax. Syntax rules are just expressed by plain C macros. The runtime routine does not dynamically allocate memory regions for memoization. Instead, statically allocated arrays are used as memoization cache tables. Therefore, programmers can implement practical parsers with CPEG, which does not depend on any specific memory management features, requiring fixed-sized memory (except for input string). To enhance usability, a translator to CPEG from an external DSL is provided, as well as a tuning mechanism to control memoization parameters. Parsing time compared to other systems when parsing JavaScript Object Notation and Java source files are given. The experimental results indicate that the performance of CPEG is competitive with other libraries.

    Download PDF (1459K)
  • Ko Sakai, Yutaka Sato
    Type: Regular Papers
    Subject area: Information Mathematics
    2018 Volume 26 Pages 345-349
    Published: 2018
    Released: April 15, 2018
    JOURNALS FREE ACCESS

    In this paper we propose a method for constructing a Gray map for a group. In an earlier paper, we suggested a new design principle of Gray maps for groups and tried to apply it to several concrete groups. Though the trial had some success, the method is not very constructive. In this paper we try to design a more constructive method based on the semidirect-product structure of the target group.

    Download PDF (166K)
  • Keisuke Furumoto, Masakatu Morii
    Type: Regular Papers
    Subject area: Information Mathematics
    2018 Volume 26 Pages 350-357
    Published: 2018
    Released: April 15, 2018
    JOURNALS FREE ACCESS

    In recent years, QR code is used in various situations such as in medical prescription and the boarding procedure at the airport in addition to conventional applications. With expanding application, there are some QR codes that cannot be read by the conventional method. In this paper, we refer to this kind of QR code as obfuscated QR code. Considering the situation that the QR code is more common, improving the encoding method is not realistic. Hence, in reading obfuscated QR code, improving only the decoding method is required. In general, Euclidean decoding is used in the decoding method of QR code. On the other hand, Generalized Minimum Distance (GMD) decoding has been proposed. GMD decoding is a method of approximately performing maximum likelihood decoding by using information of the likelihood of each symbol called reliability. However, a method for calculating the reliability information of each symbol of the two-dimensional code has not been proposed. In this paper, we propose a method of calculating reliability information using the graphical features of QR code. Then, we show the proposed method is more useful in recognition accuracy of an obfuscated QR code than the conventional method using Euclidean decoding.

    Download PDF (1290K)
  • Chuzo Iwamoto, Tatsuaki Ibusuki
    Type: Regular Papers
    Subject area: Computational Theory
    2018 Volume 26 Pages 358-361
    Published: 2018
    Released: April 15, 2018
    JOURNALS FREE ACCESS

    Dosun-Fuwari is one of Nikoli's pencil puzzles, which is played on a rectangular grid of cells. Some of the cells are colored black, and the remaining cells are divided into rooms. The purpose of the puzzle is to place balloons and iron balls according to the following rules: Place one balloon and one iron ball in each room. Balloons (resp. iron balls) are light and float (heavy and sink), so they must be placed in the top (bottom) row of the grid of cells, or in a cell right under (over) a black cell or right under other balloons (over other iron balls). It is shown that deciding whether a Dosun-Fuwari puzzle has a solution is NP-complete.

    Download PDF (311K)
  • Toshiyuki Hagiya, Toshiharu Horiuchi, Tomonori Yazaki, Tatsuya Kawahar ...
    Type: Regular Papers
    Subject area: Human-Interface Basics
    2018 Volume 26 Pages 362-374
    Published: 2018
    Released: April 15, 2018
    JOURNALS FREE ACCESS

    Many older adults are interested in smartphones. However, most of them encounter difficulties in self-instruction and need support. Text entry, which is essential for various applications, is one of the most difficult operations to master. In this paper, we propose Typing Tutor, an individualized tutoring system for text entry that detects input stumbles using a statistical approach and provides instructions. By conducting two user studies, we clarify the common difficulties that novice older adults experience and how skill level is related to input stumbles with a 12-key layout for Japanese. Based on the study, we develop Typing Tutor to support learning how to enter text on a smartphone. A two-week evaluation experiment with novice older adults (65+) showed that Typing Tutor was effective in improving their text entry proficiency, especially in the initial stage of use. In addition, we demonstrate the applicability of Typing Tutor to other keyboards and languages with the QWERTY layout for Japanese and English.

    Download PDF (2074K)
  • Hidemi Fukada
    Type: Special Issue of Information Systems
    2018 Volume 26 Pages 375
    Published: 2018
    Released: May 15, 2018
    JOURNALS FREE ACCESS
    Download PDF (37K)
  • Syunji Yazaki, Hideaki Tsuchiya, Hiroaki Ishihata
    Type: Special Issue of Information Systems
    Subject area: Network Services
    2018 Volume 26 Pages 376-385
    Published: 2018
    Released: May 15, 2018
    JOURNALS FREE ACCESS

    In this paper, we describe the details of the design and implementation of our Front-end Environment for Hands-on Activities (FEHA), which is a web-based programming environment. FEHA provides a programming environment on the web and utilizes existing Unix-like systems that equip a specialized programming environment as the build and runtime platform. FEHA controls the existing systems by using Secure SHell (SSH) and Rsync without any modification of the existing systems. We discuss a case study of FEHA in which it was applied to actual programming lectures at a university. In the lectures, 70% of the students completed registrations to use FEHA in about 3 min. In addition, they could understand how to use the FEHA and started submitting codes within several minutes after the registration. The case study shows that FEHA is able to provide a specialized programming environment for more than 100 students with a small amount of effort from the instructor and system administrator.

    Download PDF (6655K)
  • Taku Ito, Yusuke Fukazawa, Dandan Zhu, Jun Ota
    Type: Special Issue of Information Systems
    Subject area: Web Intelligence
    2018 Volume 26 Pages 386-395
    Published: 2018
    Released: May 15, 2018
    JOURNALS FREE ACCESS

    In this paper, we investigate the impact of weather context on the process of choosing foods. We mine social media on the web to create datasets on food choice, and associate food selections with weather context gathered from governmental meteorological information. From the dataset, we find that not only weather but also food events or social events on special days significantly impacts food choice. Accordingly, we propose a topic model that include the event class to represent the relationship between weather context and food choice. We quantitatively evaluated the model by perplexity, and discovered that considering both weather and event context improves prediction performance. Perplexity of the proposed model (weather and event context-aware topic model with separate topics) is (4663.0), which beats the benchmark model (4943.4). An analysis shows that combining contexts in the topic generation process yields better results that combining contexts in the word generation process. We also conduct a qualitative evaluation on the learned topic and associated foods.

    Download PDF (526K)
  • Naoki Kudo, Toshihiro Yamauchi, Thomas H. Austin
    Type: Regular Papers
    Subject area: System Security
    2018 Volume 26 Pages 396-405
    Published: 2018
    Released: May 15, 2018
    JOURNALS FREE ACCESS

    Hybrid application frameworks such as Cordova are more and more popular to create platform-independent applications (apps) because they provide special APIs to access device resources in a platform-agonistic way. By using these APIs, hybrid apps can access device resources through JavaScript. In this paper, we present a novel app-repackaging attack that repackages hybrid apps with malicious code; this code can exploit Cordova's plugin interface to steal and tamper with device resources. We address this attack and cross-site scripting attacks against hybrid apps. Since these attacks need to use plugins to access device resources, we refer to both of these attacks as Cordova plugin attacks. We further demonstrate a defense against Cordova plugin attacks through the use of a novel runtime access control mechanism that restricts access based on the mobile user's judgement. Our mechanism is easy to introduce to existing Cordova apps, and allows developers to produce apps that are resistant to Cordova plugin attacks. Moreover, we evaluate the effectiveness and performance of our mechanism.

    Download PDF (1543K)
  • Raj Dabre, Fabien Cromieres, Sadao Kurohashi
    Type: Regular Papers
    Subject area: Natural Language Processing
    2018 Volume 26 Pages 406-415
    Published: 2018
    Released: May 15, 2018
    JOURNALS FREE ACCESS

    In this paper, we explore a simple approach for “Multi-Source Neural Machine Translation” (MSNMT) which only relies on preprocessing a N-way multilingual corpus without modifying the Neural Machine Translation (NMT) architecture or training procedure. We simply concatenate the source sentences to form a single, long multi-source input sentence while keeping the target side sentence as it is and train an NMT system using this preprocessed corpus. We evaluate our method in resource poor as well as resource rich settings and show its effectiveness (up to 4 BLEU using 2 source languages and up to 6 BLEU using 5 source languages) and compare them against existing approaches. We also provide some insights on how the NMT system leverages multilingual information in such a scenario by visualizing attention. We then show that this multi-source approach can be used for transfer learning to improve the translation quality for single-source systems without using any additional corpora thereby highlighting the importance of multilingual-multiway corpora in low resource scenarios. We also extract and evaluate a multilingual dictionary by a method that utilizes the multi-source attention and show that it works fairly well despite its simplicity.

    Download PDF (506K)
feedback
Top