Vol. 44, No. 12,
Dec. 2017
Digital Library
Dynamic Directory Table: On-Demand Allocation of Directory Entries for Active Shared Cache Blocks
http://doi.org/10.5626/JOK.2017.44.12.1245
In this study we present a novel directory architecture that can dynamically allocate a directory entry for a cache block on demand at runtime only when the block is shared by more than one core. Thus, we do not maintain coherence for private blocks, substantially reducing the number of directory entries. Even for shared blocks, we allocate directory entry dynamically only when the block is actively shared, further reducing the number of directory entries at runtime. For this, we propose a new directory architecture called dynamic directory table (DDT), which is implemented as a cache of active directory entries. Through our detailed simulation on PARSEC benchmarks, we show that DDT can outperform the expensive full-map directory by a slight margin with only 17.84% of directory area across a variety of different workloads. This is achieved by its faster access and high hit rates in the small directory. In addition, we demonstrate that even smaller DDTs can give comparable or higher performance compared to recent directory optimization schemes such as SPACE and DGD with considerably less area.
Speakers’ Intention Analysis Based on Partial Learning of a Shared Layer in a Convolutional Neural Network
http://doi.org/10.5626/JOK.2017.44.12.1252
In dialogues, speakers’ intentions can be represented by sets of an emotion, a speech act, and a predicator. Therefore, dialogue systems should capture and process these implied characteristics of utterances. Many previous studies have considered such determination as independent classification problems, but others have showed them to be associated with each other. In this paper, we propose an integrated model that simultaneously determines emotions, speech acts, and predicators using a convolution neural network. The proposed model consists of a particular abstraction layer, mutually independent informations of these characteristics are abstracted. In the shared abstraction layer, combinations of the independent information is abstracted. During training, errors of emotions, errors of speech acts, and errors of predicators are partially back-propagated through the layers. In the experiments, the proposed integrated model showed better performances (2%p in emotion determination, 11%p in speech act determination, and 3%p in predicator determination) than independent determination models.
A Model for Nowcasting Commodity Price based on Social Media Data
(Jaewoo Kim, Meeyoung Cha, Jong Gun Lee
http://doi.org/10.5626/JOK.2017.44.12.1258
Capturing real-time daily information on food prices is invaluable to help policymakers and development organizations address food security problems and improve public welfare. This study analyses the possible use of large-scale online data, available due to growing Internet connectivity in developing countries, to provide updates on food security landscape. We conduct a case study of Indonesia to develop a time-series prediction model that nowcasts daily food prices for four types of food commodities that are essential in the region: beef, chicken, onion and chilli. By using Twitter price quotes, we demonstrate the capability of social data to function as an affordable and efficient proxy for traditional offline price statistics.
Statistical Ranking Recommendation System of Hangul-to-Roman Conversion for Korean Names
Jung-Hun Lee, Minho Kim, Hyuk-Chul Kwon
http://doi.org/10.5626/JOK.2017.44.12.1269
This paper focuses on the Hangul-to-roman conversion of Korean names. The proposed method recognizes existing notation and provides results according to the frequency of use. There are two main reasons for the diversity in Hangul-to-roman name conversion. The first is the indiscreet use of varied notation made domestically and overseas. The second is the customary notation of current notation. For these reasons, it has become possible to express various Roman characters in Korean names. The system constructs and converts data from 4 million people into a statistical dictionary. In the first step, the person"s name is judged through a process matching the last name. In the second step, the first name is compared and converted in the statistical dictionary. In the last step, the syllables in the name are compared and converted, and the results are ranked according to the frequency of use. This paper measured the performance compared to the existing service systems on the web. The results showed a somewhat higher performance than other systems.
A Transfer Learning Method for Solving Imbalance Data of Abusive Sentence Classification
http://doi.org/10.5626/JOK.2017.44.12.1275
The supervised learning approach is suitable for classification of insulting sentences, but pre-decided training sentences are necessary. Since a Character-level Convolution Neural Network is robust for each character, so is appropriate for classifying abusive sentences, however, has a drawback that demanding a lot of training sentences. In this paper, we propose transfer learning method that reusing the trained filters in the real classification process after the filters get the characteristics of offensive words by generated abusive/normal pair of sentences. We got higher performances of the classifier by decreasing the effects of data shortage and class imbalance. We executed experiments and evaluations for three datasets and got higher F1-score of character-level CNN classifier when applying transfer learning in all datasets.
Investigating the Feature Collection for Semantic Segmentation via Single Skip Connection
http://doi.org/10.5626/JOK.2017.44.12.1282
Since the study of deep convolutional neural network became prevalent, one of the important discoveries is that a feature map from a convolutional network can be extracted before going into the fully connected layer and can be used as a saliency map for object detection. Furthermore, the model can use features from each different layer for accurate object detection: the features from different layers can have different properties. As the model goes deeper, it has many latent skip connections and feature maps to elaborate object detection. Although there are many intermediate layers that we can use for semantic segmentation through skip connection, still the characteristics of each skip connection and the best skip connection for this task are uncertain. Therefore, in this study, we exhaustively research skip connections of state-of-the-art deep convolutional networks and investigate the characteristics of the features from each intermediate layer. In addition, this study would suggest how to use a recent deep neural network model for semantic segmentation and it would therefore become a cornerstone for later studies with the state-of-the-art network models.
Hybrid Word-Character Neural Network Model for the Improvement of Document Classification
http://doi.org/10.5626/JOK.2017.44.12.1290
Document classification, a task of classifying the category of each document based on text, is one of the fundamental areas for natural language processing. Document classification may be used in various fields such as topic classification and sentiment classification. Neural network models for document classification can be divided into two categories: word-level models and character-level models that treat words and characters as basic units respectively. In this study, we propose a neural network model that combines character-level and word-level models to improve performance of document classification. The proposed model extracts the feature vector of each word by combining information obtained from a word embedding matrix and information encoded by a character-level neural network. Based on feature vectors of words, the model classifies documents with a hierarchical structure wherein recurrent neural networks with attention mechanisms are used for both the word and the sentence levels. Experiments on real life datasets demonstrate effectiveness of our proposed model.
Rate Control Scheme for Improving Quality of Experience in the CoAP-based Streaming Environment
Hyunsoo Kang, Jiwoo Park, Kwangsue Chung
http://doi.org/10.5626/JOK.2017.44.12.1296
Recently, as the number of Internet of Things users has increased, IETF (Internet Engineering Task Force) has released the CoAP (Constrained Application Protocol). So Internet of Things have been researched actively. However, existing studies are difficult to adapt to streaming service due to low transmission rate that result from buffer underflow. In other words, one block is transmitted one block to client’s one request according to the internet environment of limited resources. The proposed scheme adaptively adjusts the rate of CON(Confirmable) message among all messages for predicting the exact network condition. Based on this, the number of blocks is determined by using buffer occupancy rate and content download rate. Therefore it improves the quality of user experience by mitigating playback interruption. Experimental results show that the proposed scheme solves the buffer underflow problem in Internet of Things streaming environment by controlling transmission rate according to the network condition.
Automated Code Generation Framework for Industrial Automation Applications based on Timed Automata Model
Kyunghyun Lee, Ikhwan Kim, Taehyoun Kim
http://doi.org/10.5626/JOK.2017.44.12.1307
Due to their convergence with state-of-the-art ICT technologies, the complexity and reliability demands of industrial automation systems have been rapidly increasing. In recent years, to cope with these demands, several research works have been carried out to adopt formal methods to the application development cycle at the early design stage. In this paper, we propose an automated code generation framework for industrial automation applications, based on a timed automata model. As a case study, we developed a formal model for a traffic light control system and verified the timing properties described in the specification. We finally demonstrated that the operation of a test-bed based on the auto-generated native code was identical to that of the model specification.
Binary Visual Word Generation Techniques for A Fast Image Search
http://doi.org/10.5626/JOK.2017.44.12.1313
Aggregating local features in a single vector is a fundamental problem in an image search. In this process, the image search process can be speeded up if binary features which are extracted almost two order of magnitude faster than gradient-based features are utilized. However, in order to utilize the binary features in an image search, it is necessary to study the techniques for clustering binary features to generate binary visual words. This investigation is necessary because traditional clustering techniques for gradient-based features are not compatible with binary features. To this end, this paper studies the techniques for clustering binary features for the purpose of generating binary visual words. Through experiments, we analyze the trade-off between the accuracy and computational efficiency of an image search using binary features, and we then compare the proposed techniques. This research is expected to be applied to mobile applications, real-time applications, and web scale applications that require a fast image search.
A Best View Selection Method in Videos of Interested Player Captured by Multiple Cameras
Hotak Hong, Gimun Um, Jongho Nang
http://doi.org/10.5626/JOK.2017.44.12.1319
In recent years, the number of video cameras that are used to record and broadcast live sporting events has increased, and selecting the shots with the best view from multiple cameras has been an actively researched topic. Existing approaches have assumed that the background in video is fixed. However, this paper proposes a best view selection method for cases in which the background is not fixed. In our study, an athlete of interest was recorded in video during motion with multiple cameras. Then, each frame from all cameras is analyzed for establishing rules to select the best view. The frames were selected using our system and are compared with what human viewers have indicated as being the most desirable. For the evaluation, we asked each of 20 non-specialists to pick the best and worst views. The set of the best views that were selected the most coincided with 54.5% of the frame selection using our proposed method. On the other hand, the set of views most selected as worst through human selection coincided with 9% of best view shots selected using our method, demonstrating the efficacy of our proposed method.
Analysis of Data Imputation in Recommender Systems
http://doi.org/10.5626/JOK.2017.44.12.1333
Recommender systems (RS) that predict a set of items a target user is likely to prefer have been extensively studied in academia and have been aggressively implemented by many companies such as Google, Netflix, eBay, and Amazon. Data imputation alleviates the data sparsity problem occurring in recommender systems by inferring missing ratings and adding them to the original data. In this paper, we point out the drawbacks of existing approaches and make suggestions for data imputation techniques. We also justify our suggestions through extensive experiments.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr