Vol. 48, No. 10,
Oct. 2021
Digital Library
MPI Progress Engine for Energy Efficient Intra-Node Point-to-Point Communication
http://doi.org/10.5626/JOK.2021.48.10.1069
The communication completion layer called progress engine in the Message Passing Interface (MPI) library recognizes changes in communication states, such as message arrivals, by polling. Although polling provides a low communication latency, its use results in low energy efficiency because the progress engine occupies CPU resources while performing polling. The decrease in energy efficiency induced by polling has become more severe as the skew has increased with the advent of exascale systems. In this paper, we suggest a progress engine that uses both polling and signal, with the Eager protocol for small message processing and the Rendezvous protocol for large message processing to perform energy-efficient intra-node communication. The OSU microbenchmark and NAS performance measurements show that the suggested signal-based progress engine improves energy efficiency as the skew increases and reduces the execution time of applications when CPU resources are shared between multiple processes.
Improving the Performance and Usability of Server Systems through Dynamic Use of Persistent Memory
http://doi.org/10.5626/JOK.2021.48.10.1077
Persistent Memory (PM) has been studied assuming that it can only be used as main memory, storage, or storage cache. In this paper, we present HMMS (Hybrid Memory Management System), which is a memory management system that allows PM to play all these roles, not exclusively, but simultaneously. Specifically, HMMS dynamically and appropriately allocates these three roles by reflecting the state of the computing system and the users’ requests. With HMMS, we aim to improve the functionality and performance of computing systems where DRAM and PM coexist.
Kor-Eng NMT using Symbolization of Proper Nouns
Myungjin Kim, Junyeong Nam, Heeseok Jung, Heeyoul Choi
http://doi.org/10.5626/JOK.2021.48.10.1084
There is progress in the field of neural machine translation, but there are cases where the translation of sentences containing proper nouns, such as, names, new words, and words that are used only within a specific group, is not accurate. To handle such cases, this paper uses the Korean-English proper noun dictionary and the symbolization method in addition to the recently proposed translation model, Transformer Model. In the proposed method, some of the words in the sentences used for learning are symbolized using a proper noun dictionary, and the translation model is trained with sentences including the symbolized words. When translating a new sentence, the translation is completed by symbolizing, translation, and desymbolizing. The proposed method was compared with a model without symbolization, and for some cases improvement was quantitatively confirmed with the BLEU score. In addition, several examples of translation were also presented along with commercial service results.
A Weight-based Multi-domain Recommendation System for Alleviating the Cold-Start Problem
http://doi.org/10.5626/JOK.2021.48.10.1090
A recommendation system predicts users’ ratings based on users’ past behaviors and item preferences. One of the most famous types of recommendation systems is the collaborative filtering method that predicts users’ ratings based on the rating information from users with similar preferences. In order to predict the preferences of users, we need adequate information about users’ interactive information on items. Otherwise, it is very difficult to make accurate predictions for users without adequate information. This phenomenon is called the cold-start problem. In this paper, we propose a multi-domain recommendation system that utilizes the rating information of multiple domains. We propose a method that calculates the weight of each auxiliary domain to maximize the confidence of predicted ratings from multiple auxiliary domains and verify the performance of the proposed method through extensive experiments. As a result, we demonstrate that our algorithm produces better recommendation results compared to the classical algorithms simply utilized in multiple domain settings.
Analyzing the Impact of Sequential Context Learning on the Transformer Based Korean Text Summarization Model
Subin Kim, Yongjun Kim, Junseong Bang
http://doi.org/10.5626/JOK.2021.48.10.1097
Text summarization reduces the sequence length while maintaining the meaning of the entire article body, solving the problem of overloading information and helping readers consume information quickly. To this end, research on a Transformer-based English text summarization model has been actively conducted. Recently, an abstract text summary model reflecting the characteristics of English with a fixed word order by adding a Recurrent Neural Network (RNN)-based encoder was proposed. In this paper, we study the effect of sequential context learning on the text abstract summary model by using an RNN-based encoder for Korean, which has more free word order than English. Transformer-based model and a model that added RNN-based encoder to existing Transformer model are trained to compare the performance of headline generation and article body summary for the Korean articles collected directly. Experiments show that the model performs better when the RNN-based encoder is added, and that sequential contextual information learning is required for Korean abstractive text summarization.
Survey of EEG Neurofeedback methods for Attention Improvement
Hyunji Kim, Daeun Gwon, Kyungho Won, Sung Chan Jun, Minkyu Ahn
http://doi.org/10.5626/JOK.2021.48.10.1105
Neurofeedback is a method through which a user self-regulates the brain state using the feedback of his/her own brain signal. This can be used to restore or improve brain functions. In this study, we reviewed 108 articles on electroencephalogram (EEG) neurofeedback for attention improvement and surveyed the important parameters. As a result, we found that most studies were conducted with patient subjects and mostly brain signals were recorded from central areas on the scalp by using wet and wire EEG systems. Sensory-motor-rhythm or the ratio between theta and low beta rhythms were used as attention index, and this information was provided to users through auditory or visual stimuli. In addition, Continuous Performance Test or Go/NoGo test was employed for behavior evaluation. Based on these results, we suggest the following directions for the further advancement of the practical neurofeedback system; the future work should target non-patient subjects and utilize wireless/dry EEG devices and virtual/augmented reality for increasing user convenience and building more immersive application. Lastly, a standard or guide for developing usable neurofeedback applications should be established.
SMERT: Single-stream Multimodal BERT for Sentiment Analysis and Emotion Detection
Kyeonghun Kim, Jinuk Park, Jieun Lee, Sanghyun Park
http://doi.org/10.5626/JOK.2021.48.10.1122
Sentiment Analysis is defined as a task that analyzes subjective opinion or propensity and, Emotion Detection is the task that finds emotions such as ‘happy’ or ‘sad’ from text data. Multimodal data refers to the appearance of image and voice data in addition to text data. In prior research, RNN or cross-transformer models were used, however, RNN models have long-term dependency problems. Also, since cross-transformer models could not capture the attribute of modalities, they got worse results. To solve those problems, we propose SMERT based on a single-stream transformer ran on a single network. SMERT can get joint representation for Sentiment Analysis and Emotion Detection. Besides, we use BERT tasks which are improved to utilize for multimodal data. To present the proposed model, we verify the superiority of SMERT through a comparative experiment on the combination of modalities using the CMU-MOSEI dataset and various evaluation metrics.
Design and Evaluation of Loss Functions based on Classification Models
Hyun-Kyu Jeon, Yun-Gyung Cheong
http://doi.org/10.5626/JOK.2021.48.10.1132
Paraphrase generation is a task in which the model generates an output sentence conveying the same meaning as the given input text but with a different representation. Recently, paraphrase generation has been widely used for solving the task of using artificial neural networks with supervised learning between the model’s prediction and labels. However, this method gives limited information because it only detects the representational difference. For that reason, we propose a method to extract semantic information with classification models and use them for the training loss function. Our evaluations showed that the proposed method outperformed baseline models.
Partially Collective Spatial Keyword Query Processing Based on Spatial Keyword Similarity
Ah Hyun Lee, Sehwa Park, Seog Park
http://doi.org/10.5626/JOK.2021.48.10.1142
Collective spatial keyword queries return Points of Interest (POI), which are close to the query location and contain all the presented set of keywords. However, existing studies only consider a fixed number of query keywords, which is not adequate to satisfy the user. They do not care about the preference of a partial keyword set, and a flexible keyword set needs to be selected for the preference of each POI. We thus propose a new query, called Partially Collective Spatial Keyword Query, which flexibly considers keywords that fit the preference for each POI. Since this query is a combinatorial optimization problem, the query processing time increases rapidly as the number of POIs increases. Therefore, to address these problems, we propose a keyword-based search technique that reduces the overall search space. Furthermore, we propose heuristic techniques, which include the linear search-based terminal node pruning technique, approximation algorithm, and threshold-based pruning technique.
Distributed Processing of Deep Learning Inference Models for Data Stream Classification
Hyojong Moon, Siwoon Son, Yang-Sae Moon
http://doi.org/10.5626/JOK.2021.48.10.1154
The increased generation of data streams has subsequently led to increased utilization of deep learning. In order to classify data streams using deep learning, we need to execute the model in real-time through serving. Unfortunately, the serving model incurs long latency due to gRPC or HTTP communication. In addition, if the serving model uses a stacking ensemble method with high complexity, a longer latency occurs. To solve the long latency challenge, we proposed distributed processing solutions for data stream classification using Apache Storm. First, we proposed a real-time distributed inference method based on Apache Storm to reduce the long latency of the existing serving method. The present study"s experimental results showed that the proposed distributed inference method reduces the latency by up to 11 times compared to the existing serving method. Second, to reduce the long latency of the stacking-based inference model for detecting malicious URLs, we proposed four distributed processing techniques for classifying URL streams in real-time. The proposed techniques are Independent Stacking, Sequential Stacking, Semi-Sequential Stacking, and Stepwise-Independent Stacking. Our study experimental results showed that Stepwise-Independent Stacking, whose characteristics are similar to those of independent execution and sequential processing, is the best technique for classifying URL streams with the shortest latency.
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