• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.6
• /
• pp.773-778
• /
• 2009

This paper studied the training methods less affected by the emotional variation for the development of the robust speech recognition system. For this purpose, the effect of emotional variations on the speech signal were studied using speech database containing various emotions. The performance of the speech recognition system trained by using the speech signal containing no emotion is deteriorated if the test speech signal contains the emotions because of the emotional difference between the test and training data. In this study, it is observed that vocal tract length of the speaker is affected by the emotional variation and this effect is one of the reasons that makes the performance of the speech recognition system worse. In this paper, vocal tract normalization method is used to develop the robust speech recognition system for emotional variations. Experimental results from the isolated word recognition using HMM showed that the vocal tract normalization method reduced the error rate of the conventional recognition system by 41.9% when emotional test data was used.

• Smart Media Journal
• /
• v.7 no.4
• /
• pp.17-23
• /
• 2018

We proposed a LSTM-based RNN model that can effectively perform the automatic spacing characteristics. For those long or noisy sentences which are known to be difficult to handle within Neural Network Learning, we defined a proper input data format and decoding data format, and added dropout, bidirectional multi-layer LSTM, layer normalization, and attention mechanism to improve the performance. Despite of the fact that Sejong corpus contains some spacing errors, a noise-robust learning model developed in this study with no overfitting through a dropout method helped training and returned meaningful results of Korean word spacing and its patterns. The experimental results showed that the performance of LSTM sequence-to-sequence model is 0.94 in F1-measure, which is better than the rule-based deep-learning method of GRU-CRF.

• Proceedings of the Computational Structural Engineering Institute Conference
• /
• 2010.04a
• /
• pp.687-691
• /
• 2010

구조물의 건전도를 평가하기 위해 상시 구조물 계측을 이용한 Structural Health Monitoring (SHM) 시스템을 적용하게 된다. SHM 시스템의 궁극적 목적은 계측된 데이터를 이용하여 구조물의 손상위치 및 손상정도를 분석하여 거주자에게 유지관리정보와 대처요령 신속하게 제공하는 것이다. 따라서 본 연구에서는 구조물의 손상탐지를 위해 인공신경망(Artificial Neural Network)을 도입한 알고리즘을 수립하고, 이를 3층 실대 RC Mock-up 구조물에 적용하여 성능을 평가하였다. 먼저 인공신경망의 학습을 위해 구조해석 프로그램을 이용하여 구조물의 손상에 따른 동적특성 변화 데이터베이스를 구축하였다. 그리고 학습된 인공망에 실제 구조물에서 추출한 동특성의 변화를 입력하여 손상탐지를 실시하였다. 이를 통해 인공신경망의 학습방법, 학습데이터의 정규화 방법 등을 규명하고 인공신경망을 이용한 손상탐지의 효과를 분석하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2008.11a
• /
• pp.192-195
• /
• 2008

본 논문에서는 발전하는 3D 그래픽스 기술을 이용하여 문화재의 도면 실루엣을 생성하는 방법을 제안하고자 한다. 3D 스캐너로 정밀 실측된 3D 데이터를 이용하여 문화재의 도면을 생성하기 위한 벡터 실루엣(Silhouette) 추출 과정은 다음과 같다. 먼저 실측된 3D 데이터를 정규화 된 3D공간으로 이동하고, 이동 후에는 데이터에 존재하는 모든 에지(edge)를 검출하여 에지리스트(edge list)를 생성한다. 생성된 에지리스트는 다시 윤곽에지(Contour edge)와 주름에지(Crease edge)로 분류하는데, 윤곽에지는 문화재의 윤곽 실루엣을 형성하는데 이용하고, 윤곽에지를 제외한 주름에지는 문화재의 표면 특징을 나타내는 내부문양 실루엣을 형성하는데 이용한다. 내부문양 실루엣은 사용자가 입력하는 임계값과 주름에지를 구성하는 두면의 방향 벡터의 내적을 비교하여 추출한다. 추출한 벡터 실루엣은 윤곽 실루엣과 내부문양 실루엣으로 구분되며, 두 벡터 실루엣을 이용함으로써 문화재의 구조적 해석과 표면의 특징을 해석할 수 있는 도면 실루엣 생성이 가능했다.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.44 no.6 s.360
• /
• pp.42-49
• /
• 2007

CABAC(Context-based Adaptive Binary Arithmetic Coding) among various entropy coding schemes which are used to improve compression efficiency in H.264/AVC has a high hardware complexity and the fast calculation is difficult because data dependancy exists in the bit-serial process. In this paper, the proposed architecture efficiently compose the renormalization process of binary arithmetic encoder which is an important part of CABAC used in H.264/AVC. At every clock cycle, the input symbol is encoded regardless of the iteration of the renormalization process for every input symbol. Also, the proposed architecture can deal with the bitsOutstanding up to 127 which is adopted to handle the carry generation problem and encode input symbol without stall. The proposed architecture with three-stage pipeline has been synthesized using the 0.18um Dongbu-Anam standard cell library and can be operated at 290MHz.

• Knowledge Management Research
• /
• v.22 no.4
• /
• pp.27-40
• /
• 2021

Click-Through Rate (CTR) prediction is a key function that determines the ranking of candidate items in the recommendation system and recommends high-ranking items to reduce customer information overload and achieve profit maximization through sales promotion. The fields of natural language processing and image classification are achieving remarkable growth through the use of deep neural networks. Recently, a transformer model based on an attention mechanism, differentiated from the mainstream models in the fields of natural language processing and image classification, has been proposed to achieve state-of-the-art in this field. In this study, we present a method for improving the performance of a transformer model for CTR prediction. In order to analyze the effect of discrete and categorical CTR data characteristics different from natural language and image data on performance, experiments on embedding regularization and transformer normalization are performed. According to the experimental results, it was confirmed that the prediction performance of the transformer was significantly improved when the L2 generalization was applied in the embedding process for CTR data input processing and when batch normalization was applied instead of layer normalization, which is the default regularization method, to the transformer model.

• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.7 no.11
• /
• pp.71-79
• /
• 2017

In this paper, we implemented the exchange rate forecasting neural network using heterogeneous computing. Exchange rate forecasting requires a large amount of data. We used a neural network that could leverage this data accordingly. Neural networks are largely divided into two processes: learning and verification. Learning took advantage of the CPU. For verification, RTL written in Verilog HDL was run on FPGA. The structure of the neural network has four input neurons, four hidden neurons, and one output neuron. The input neurons used the US $ 1, Japanese 100 Yen, EU 1 Euro, and UK £ 1. The input neurons predicted a Canadian dollar value of $ 1. The order of predicting the exchange rate is input, normalization, fixed-point conversion, neural network forward, floating-point conversion, denormalization, and outputting. As a result of forecasting the exchange rate in November 2016, there was an error amount between 0.9 won and 9.13 won. If we increase the number of neurons by adding data other than the exchange rate, it is expected that more precise exchange rate prediction will be possible.

• Proceedings of the Korean Society for Emotion and Sensibility Conference
• /
• 2009.11a
• /
• pp.239-242
• /
• 2009

본 연구에서는 독립차원 감성추론이 가능한 추론엔진과 하드웨어를 모바일 상황에서 사용가능 하도록 설계하였다. 설계된 시스템은 크게 감성 신호 센싱 디바이스에서 수신한 생체 신호 데이터를 이용해 독립차원의 감성을 판단하는 모바일 기반 감성 추론 엔진 소프트웨어 시스템과, 감성 추론 엔진을 탑재하고 감성 신호 센싱 디바이스와 최종 감성 응용 서비스 단말과의 인터페이스 기능을 갖는 모바일 감성 단말 하드웨어 시스템으로 구성된다. 모바일 독립차원 감성 추론 엔진은 감성 신호 센싱 디바이스로부터 측정되어 정규화(Normalization)된 감성 유발 채널별 특징 신호 데이터를 입력 받아 사용자 별 감성 추론을 위한 개인화 인자를 추출하고, 최종적으로 사용자의 현재 감성 상태를 추론하는 기능을 수행한다. 모바일 감성 단말 하드웨어 시스템은 독립차원 감성 추론 엔진을 내장해 실행하고, 감성 추론을 위한 감성 신호 센싱 디바이스와의 인터페이스와 추론된 감성 정보 데이터를 감성 증강 UI 기반 서비스 플랫폼으로 전송하기 위한 인터페이스 기능을 수행한다. 본 연구에서 설계된 시스템은 다양한 환경에서 실시간으로 감성을 추론할 수 있는 알고리즘과 하드웨어를 구축하는 기술로 향후 다양한 제품에 접목하여 다양한 감성 서비스가 가능할 것으로 예상된다.

• Journal of The Korean Association of Information Education
• /
• v.15 no.1
• /
• pp.147-154
• /
• 2011

In this Paper, we suggest the effective pre-correction algorithm on sensor values and the classification algorithm for gesture recognition that use values for each axis of the accelerometer to send data(a number or specific input data) to device. we know that creation of reliable preprocessed data in experimental results through the error rate of X-Axis and Y-Axis for pre-correction and post-correction. we can show high recognition rate through recognizer using the normalization and classification algorithm for the preprocessed data.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.22 no.3
• /
• pp.565-573
• /
• 2012

To investigate the business corruption, the obtainments of the business data such as personnel, manufacture, accounting and distribution etc., is absolutely necessary. Futhermore, the investigator should have the systematic extraction solution from the business data of the enterprise database, because most company manage each business data through the distributed database system, In the general business environment, the database exists in the system with upper layer application and big size file server. Besides, original resource data which input by user are distributed and stored in one or more table following the normalized rule. The earlier researches of the database structure analysis mainly handled the table relation for database's optimization and visualization. But, in the point of the digital forensic, the data, itself analysis is more important than the table relation. This paper suggests the extraction technique from the table relation which already defined in the database. Moreover, by the systematic analysis process based on the domain knowledge, analyzes the original business data structure stored in the database and proposes the solution to extract table which is related incident.