• Journal of KIISE
• /
• v.44 no.3
• /
• pp.306-313
• /
• 2017

Named entity recognition (NER) seeks to locate and classify named entities in text into pre-defined categories such as names of persons, organizations, locations, expressions of times, etc. Recently, many state-of-the-art NER systems have been implemented with bidirectional LSTM CRFs. Deep learning models based on long short-term memory (LSTM) generally depend on word representations as input. In this paper, we propose an approach to expand word representation by using pre-trained word embedding, part of speech (POS) tag embedding, syllable embedding and named entity dictionary feature vectors. Our experiments show that the proposed approach creates useful word representations as an input of bidirectional LSTM CRFs. Our final presentation shows its efficacy to be 8.05%p higher than baseline NERs with only the pre-trained word embedding vector.

• Korean Journal of Biological Psychiatry
• /
• v.8 no.1
• /
• pp.156-161
• /
• 2001

We reported two cases of amantadine treatment in traumatic brain injury patients and reviewed the literature of amantadine treatment of those patients. Problems with short-term memory, attention, planning, problem solving, impulsivity, disinhibition, poor motivation, and other behavioral and cognitive deficit could occur following traumatic brain injury or other types of acquired brain injury. This report described results of amantadine using in two patients with this type of symptom profile. Patients received neuropsychiatric examination as well as BPRS and Barthel index. These patients were improved, respectively from 57 point to 82 point(case 1), from 85 to 94(case 2) in Barthel index, and from 66 point to 35 point(case 1), from 55 to 32 point(case 2) in BPRS. These two patients did not reveal any other adverse effect. The rationale for using amantadine were discussed.

• Journal of Korea Multimedia Society
• /
• v.23 no.5
• /
• pp.641-649
• /
• 2020

As the usage of mobile devices extremely increases, malicious mobile apps(applications) that target mobile users are also increasing. It is challenging to detect these malicious apps using traditional malware detection techniques due to intelligence of today's attack mechanisms. Deep learning (DL) is an alternative technique of traditional signature and rule-based anomaly detection techniques and thus have actively been used in numerous recent studies on malware detection. In order to develop DL-based defense mechanisms against intelligent malicious apps, feeding recent datasets into DL models is important. In this paper, we develop a DL-based model for detecting intelligent malicious apps using KU-CISC 2018-Android, the most up-to-date dataset consisting of benign and malicious Android apps. This dataset has hardly been addressed in other studies so far. We extract OPcode sequences from the Android apps and preprocess the OPcode sequences using an N-gram model. We then feed the preprocessed data into LSTM and apply the concept of Information Gain to improve performance of detecting malicious apps. Furthermore, we evaluate our model with numerous scenarios in order to verify the model's design and performance.

• KIISE Transactions on Computing Practices
• /
• v.23 no.5
• /
• pp.322-327
• /
• 2017

The classification problem in the field of Natural Language Processing has been studied for a long time. Continuing forward with our previous research, which classifies large-scale text using Convolutional Neural Networks (CNN), we implemented Recurrent Neural Networks (RNN), Long-Short Term Memory (LSTM) and Gated Recurrent Units (GRU). The experiment's result revealed that the performance of classification algorithms was Multinomial Naïve Bayesian Classifier < Support Vector Machine (SVM) < LSTM < CNN < GRU, in order. The result can be interpreted as follows: First, the result of CNN was better than LSTM. Therefore, the text classification problem might be related more to feature extraction problem than to natural language understanding problems. Second, judging from the results the GRU showed better performance in feature extraction than LSTM. Finally, the result that the GRU was better than CNN implies that text classification algorithms should consider feature extraction and sequential information. We presented the results of fine-tuning in deep neural networks to provide some intuition regard natural language processing to future researchers.

• Journal of Korean Neurosurgical Society
• /
• v.53 no.3
• /
• pp.139-144
• /
• 2013

Objective : Transient anterograde amnesia is occasionally observed in a number of conditions, including migraine, focal ischemia, venous flow abnormalities, and after general anesthesia. The inhalation anesthetic, isoflurane, is known to induce transient anterograde amnesia. We examined the involvement of brain-derived neurotrophic factor (BDNF) and its receptor tyrosine kinase B (TrkB) in the underlying mechanisms of the isoflurane-induced transient anterograde amnesia. Methods : Adult male Sprague-Dawley rats were divided into three groups : the control group, the 10 minutes after recovery from isoflurane anesthesia group, and the 2 hours after recovery from isoflurane anesthesia group (n=8 in each group). The rats in the isoflurane-exposed groups were anesthetized with 1.2% isoflurane in 75% nitrous oxide and 25% oxygen for 2 hours in a Plexiglas anesthetizing chamber. Short-term memory was determined using the step-down avoidance task. BDNF and TrkB expressions in the hippocampus were evaluated by immunofluorescence staining and western blot analysis. Results : Latency in the step-down avoidance task was decreased 10 minutes after recovery from isoflurane anesthesia, whereas it recovered to the control level 2 hours after isoflurane anesthesia. The expressions of BDNF and TrkB in the hippocampus were decreased immediately after isoflurane anesthesia but were increased 2 hours after isoflurane anesthesia. Conclusion : In this study, isoflurane anesthesia induced transient anterograde amnesia, and the expressions of BDNF and TrkB in the hippocampus might be involved in the underlying mechanisms of this transient anterograde amnesia.

• Molecules and Cells
• /
• v.26 no.6
• /
• pp.611-615
• /
• 2008

DNA methylation is an epigenetic mechanism for gene silencing. In Arabidopsis, MET1 is the primary DNA methyltransferase that maintains CG DNA methylation. Plants having an overall reduction of MET1 activity, caused by a met1 mutation or a constitutively expressed MET1 antisense gene, display genome hypomethylation, inappropriate gene and transposon transcription, and developmental abnormalities. However, the effect of a transient reduction in MET1 activity caused by inhibiting MET1 expression in a restricted set of cells is not known. For this reason, we generated transgenic plants with a MET1 antisense gene fused to the DEMETER (DME) promoter (DME:MET1 a/s). Here we show that DME is expressed in leaf primordia, lateral root primoridia, in the region distal to the primary root apical meristem, which are regions that include proliferating cells. Endogenous MET1 expression was normal in organs where the DME:MET1 a/s was not expressed. Although DME promoter is active only in a small set of cells, these plants displayed global developmental abnormalities. Moreover, centromeric repeats were hypomethylated. The developmental defects were accumulated by the generations. Thus, not maintaining CG methylation in a small population of proliferating cells flanking the meristems causes global developmental and epigenetic abnormalities that cannot be rescued by restoring MET1 activity. These results suggest that during plant development there is little or no short-term molecular memory for reestablishing certain patterns of CG methylation that are maintained by MET1. Thus, continuous MET1 activity in dividing cells is essential for proper patterns of CG DNA methylation and development.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.12
• /
• pp.209-216
• /
• 2019

In view of the educational effects and social changes of software education, equal opportunities for software education are needed regardless of general students and students with disabilities. However, studies on software education for general students have been actively conducted, but studies on software education for students with disabilities are insufficient. In this study, we developed a robot education software education program for students with developmental disabilities. Developing robot-enabled software education programs for students with developmental disabilities is meaningful in terms of expanding software education opportunities for all. In addition, the robot-based software education program is easy to motivate students with developmental disabilities with low task concentration, short-term memory, and low sociality. Significant changes will be made not only in terms of management capacity, but also in terms of self-efficacy and confidence.

• Journal of Korea Spatial Information System Society
• /
• v.8 no.2 s.17
• /
• pp.39-52
• /
• 2006

With the recent the use of spatio-temporal data mining which can extract various knowledge such as movement patterns of moving objects in history data of moving object gets increasing. However, the existing movement pattern extraction methods create lots of candidate movement patterns when the minimum support is low. Therefore, in this paper, we suggest the STMPE(Spatio-Temporal Movement Pattern Extraction) algorithm in order to efficiently extract movement patterns of moving objects from the large capacity of spatio-temporal data. The STMPE algorithm generalizes spatio-temporal and minimizes the use of memory. Because it produces and keeps short-term movement patterns, the frequency of database scan can be minimized. The STMPE algorithm shows more excellent performance than other movement pattern extraction algorithms with time information when the minimum support decreases, the number of moving objects increases, and the number of time division increases.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.2
• /
• pp.45-52
• /
• 2020

Since the opening of the national petition site, it has attracted much attention. In this paper, we perform topic analysis of the national petition site and propose a prediction model for answerable petitions based on deep learning. First, 1,500 petitions are collected, topics are extracted based on the petitions' contents. Main subjects are defined using K-means clustering algorithm, and detailed subjects are defined using topic modeling of petitions belonging to the main subjects. Also, long short-term memory (LSTM) is used for prediction of answerable petitions. Not only title and contents but also categories, length of text, and ratio of part of speech such as noun, adjective, adverb, verb are also used for the proposed model. Our experimental results show that the type 2 model using other features such as ratio of part of speech, length of text, and categories outperforms the type 1 model without other features.

• Journal of Digital Convergence
• /
• v.13 no.3
• /
• pp.295-304
• /
• 2015

We assessed reading-related skills in children with reading comprehension difficulties ("Poor comprehenders"), word decoding difficulties ("Poor decoders") and normal readers, matched for age and nonverbal IQ. The reading-related skill tests used in our study are phonological processing, visual processing test, and receptive vocabulary test. The authors argue that children who had difficulty in reading comprehension had lower scores only on the phonological short-term memory test compared with normal readers, although their performance on receptive vocabulary and visual processing tests are comparable to normal readers. The results of our study revealed that poor decoders had lower scores on the phonological processing, visual processing, and receptive vocabulary tests.