• 한국정보통신학회논문지
• /
• 제17권2호
• /
• pp.397-404
• /
• 2013

본 논문에서는 Adaboost알고리즘을 이용한 얼굴인식 하드웨어 시스템의 구조를 제안하였다. 제안된 하드에어 구조는 초당 30프레임을 가지며 실시간 처리가 가능하다. 또한 Adaboost알고리즘을 이용하여 얼굴 특징 데이터를 학습하였고, 영상 크기 축소부와 적분 영상 추출부 그리고 얼굴 비교부, 메모리 인터페이스부, 데이터 그룹화, 검출결과 표시부 등으로 구성되었다. 제안된 하드웨어 구조는 사이클당 1포인트를 계산 할 수 있는 구조로 속도의 향상을 가져오며 full HD($1920{\times}1080$)의 경우에는 총 사이클 수 $2,316,087{\times}30=69,482,610$로 약 70MHz의 속도를 가진다. 제안된 하드웨어 구조는 Verilog HDL로 디자인되었고, Mentor Graphics Modelsim을 이용하여 검증하였으며, 합성은 FPGA Xilinx Virtex5 XC5VLX330을 이용하여 칩의 대략 35%인 74,757 Slice LUT와 45MHz의 주파수에서 동작한다.

• 한국컴퓨터정보학회논문지
• /
• 제25권11호
• /
• pp.27-40
• /
• 2020

본 논문은 장단기기억신경망(LSTM)이 영어를 배우면서 학습한 암묵적 통사 관계인 필러-갭 의존 관계를 조사하여 영어 문장 학습량과 한국인 영어 학습자(L2ers)의 문장 처리 패턴 간의 상관관계를 규명한다. 이를 위해, 먼저 장단기기억신경망 언어모델(LSTM LM)을 구축하였다. 이 모델은 L2ers가 영어 학습 과정에서 잠재적으로 배울 수 있는 L2 코퍼스의 영어 문장들로 심층학습을 하였다. 다음으로, 이 언어 모델을 이용하여 필러-갭 의존 관계 구조를 위반한 영어 문장을 대상으로 의문사 상호작용 효과(wh-licensing interaction effect) 즉, 정보 이론의 정보량인 놀라움(surprisal)의 정도를 계산하여 문장 처리 양상을 조사하였다. 또한 L2ers 언어모델과 상응하는 원어민 언어모델을 비교 분석함으로써, 두 언어모델이 문장 처리에서 필러-갭 의존 관계에 내재된 추상적 구문 구조를 추적할 수 있음을 보여주었을 뿐만 아니라, 또한 선형 혼합효과 회귀모델을 사용하여 본 논문의 중심 연구 주제인 의존 관계 처리에 있어서 원어민 언어모델과 L2ers 언어모델간 통계적으로 유의미한 차이가 존재함을 규명하였다.

• 영어어문교육
• /
• 제12권3호
• /
• pp.267-286
• /
• 2006

The purpose of the study was to investigate what kinds of learning strategies EFL learners use to learn English grammar and what is benefit from structured grammar input processing. Students of the study consisted of 48 college students who took Practical English Grammar at a university in Kyung-Gi area and were divided into two groups based on grammar scores. The students were asked to take two grammar tasks and grammar tests and complete a survey including questions on grammar strategy and input processing. The results of the study are as follows. First, learners' grammar level has an effect on use of grammar attack strategy including asking teachers, using grammar books and given contexts whereas there was no significant difference between groups in the planning strategies, Among memory strategies, using grammar exercise and linking with already known structure demonstrated a significant difference between groups. Second, with regard to input processing, high level students got higher score on how much they understood the structured grammar input compared with low level students. Third, explicit implicit instruction added to input processing seems more comprehensible and more available than structured input only, Finally, it showed that there is positive relationship between perception and score of input processing tasks and grammar tests. Especially, learners' perception of input processing correlated more with final tests and tasks. Therefore, it suggests that the more input processing task need to develop and utilize in order to facilitate learners' intake.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권6호
• /
• pp.2188-2203
• /
• 2021

With the rapid development of mobile Internet, smart phones have been widely popularized, among which Android platform dominates. Due to it is open source, malware on the Android platform is rampant. In order to improve the efficiency of malware detection, this paper proposes deep learning Android malicious detection system based on behavior features. First of all, the detection system adopts the static analysis method to extract different types of behavior features from Android applications, and extract sensitive behavior features through Term frequency-inverse Document Frequency algorithm for each extracted behavior feature to construct detection features through unified abstract expression. Secondly, Long Short-Term Memory neural network model is established to select and learn from the extracted attributes and the learned attributes are used to detect Android malicious applications, Analysis and further optimization of the application behavior parameters, so as to build a deep learning Android malicious detection method based on feature analysis. We use different types of features to evaluate our method and compare it with various machine learning-based methods. Study shows that it outperforms most existing machine learning based approaches and detects 95.31% of the malware.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권10호
• /
• pp.3211-3229
• /
• 2022

To solve the problems existing in the process of Weibo disaster rumor recognition, such as lack of corpus, poor text standardization, difficult to learn semantic information, and simple semantic features of disaster rumor text, this paper takes Sina Weibo as the data source, constructs a dataset for Weibo disaster rumor recognition, and proposes a deep learning model BERT_AT_Stacked LSTM for Weibo disaster rumor recognition. First, add adversarial disturbance to the embedding vector of each word to generate adversarial samples to enhance the features of rumor text, and carry out adversarial training to solve the problem that the text features of disaster rumors are relatively single. Second, the BERT part obtains the word-level semantic information of each Weibo text and generates a hidden vector containing sentence-level feature information. Finally, the hidden complex semantic information of poorly-regulated Weibo texts is learned using a Stacked Long Short-Term Memory (Stacked LSTM) structure. The experimental results show that, compared with other comparative models, the model in this paper has more advantages in recognizing disaster rumors on Weibo, with an F1_Socre of 97.48%, and has been tested on an open general domain dataset, with an F1_Score of 94.59%, indicating that the model has better generalization.

• 전자통신동향분석
• /
• 제34권4호
• /
• pp.55-64
• /
• 2019

In this study, we introduce trends in neural-network-based deep learning research applied to dialogue systems. Recently, end-to-end trainable goal-oriented dialogue systems using long short-term memory, sequence-to-sequence models, among others, have been studied to overcome the difficulties of domain adaptation and error recognition and recovery in traditional pipeline goal-oriented dialogue systems. In addition, some research has been conducted on applying reinforcement learning to end-to-end trainable goal-oriented dialogue systems to learn dialogue strategies that do not appear in training corpora. Recent neural network models for end-to-end trainable chit-chat systems have been improved using dialogue context as well as personal and topic information to produce a more natural human conversation. Unlike previous studies that have applied different approaches to goal-oriented dialogue systems and chit-chat systems respectively, recent studies have attempted to apply end-to-end trainable approaches based on deep neural networks in common to them. Acquiring dialogue corpora for training is now necessary. Therefore, future research will focus on easily and cheaply acquiring dialogue corpora and training with small annotated dialogue corpora and/or large raw dialogues.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권8호
• /
• pp.4026-4040
• /
• 2019

In this paper, we present an approach for detecting slide transitions in lecture videos by introducing the spatio-temporal residual networks. Given a lecture video which records the digital slides, the speaker, and the audience by multiple cameras, our goal is to find keyframes where slide content changes. Since temporal dependency among video frames is important for detecting slide changes, 3D Convolutional Networks has been regarded as an efficient approach to learn the spatio-temporal features in videos. However, 3D ConvNet will cost much training time and need lots of memory. Hence, we utilize ResNet to ease the training of network, which is easy to optimize. Consequently, we present a novel ConvNet architecture based on 3D ConvNet and ResNet for slide transition detection in lecture videos. Experimental results show that the proposed novel ConvNet architecture achieves the better accuracy than other slide progression detection approaches.

• 한국멀티미디어학회논문지
• /
• 제23권2호
• /
• pp.241-246
• /
• 2020

Hadoop is a framework to process large data sets in a distributed way across clusters of nodes. It has been a popular platform to process big data, but in recent years, other platforms became competitive ones depending on the characteristics of the application. Spark is one of distributed platforms to enable real-time data processing and improve overall processing performance over Hadoop by introducing in-memory processing instead of disk I/O. Whereas Hadoop is designed to work on Java and data analysis is processed using Java API, Spark provides a variety of APIs with Scala, Python, Java and R. In this paper, the goal is to find out whether the APIs of different programming languages af ect the performances in Spark. We chose two popular APIs: Python and Scala. Python is easy to learn and is used in AI domain in a wide range. Scala is a programming language with advantages of parallelism. Our experiment shows much faster processing with Scala API than Python API. For the performance issues on AI-based analysis, further study is needed.

• 비교문화연구
• /
• 제37권
• /
• pp.267-286
• /
• 2014

The purpose of this study is to improve the writing skills through activities to read and summarize the internet children newspaper article. The subjects of study are the college students of A2-B1 level in the French writing classes. The range of study was as follows: - As the previous activity of writing, activities of teaching and learning of vocabularies to comprehend the internet children newspaper article. - learn about the rules of summary - writing the summary The children's newspaper used in this study has the advantage that can increase the learning motivation of learners as having a topicality by itself and a level of easy language. The summary activities can be called a comprehensive activities of teaching and learning that combine the critical reading ability that can distinguish important information and secondary one with the creative writing ablility that can reconstruct one's own style from the selected content. In addition, the summary assists the understanding of a text and is a help to its memory. It is the strategy of reading comprehension and also is simultaneously the strategy of writing that can write with one's own vocabulary by newly structuring the text. The results of this study will provide a vitality for the education environment and field of study of French language that have been neglected the writing ability. Moreover it will be the motivation to propose a way of a balanced French language communication to our French language learners weighted on oral communication.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권5호
• /
• pp.1590-1609
• /
• 2021

Laughter is one of the most important nonverbal sound that human generates. It is a means for expressing his emotions. The acoustic and contextual features of this specific sound are different from those of speech and many difficulties arise during their modeling process. During this work, we propose an audio laughter generation system based on unsupervised generative models: the autoencoder (AE) and its variants. This procedure is the association of three main sub-process, (1) the analysis which consist of extracting the log magnitude spectrogram from the laughter database, (2) the generative models training, (3) the synthesis stage which incorporate the involvement of an intermediate mechanism: the vocoder. To improve the synthesis quality, we suggest two hybrid models (LSTM-VAE, GRU-VAE and CNN-VAE) that combine the representation learning capacity of variational autoencoder (VAE) with the temporal modelling ability of a long short-term memory RNN (LSTM) and the CNN ability to learn invariant features. To figure out the performance of our proposed audio laughter generation process, objective evaluation (RMSE) and a perceptual audio quality test (listening test) were conducted. According to these evaluation metrics, we can show that the GRU-VAE outperforms the other VAE models.