• The Korean Journal of Petroleum Geology
• /
• v.5 no.1_2 s.6
• /
• pp.36-47
• /
• 1997

This study presents multi-variable sequence model for a broader application of sequence concept proposed by Exxon group. The concept of the multi-variable model is based on the fact that internal organization and boundary type of the sequences are determined by three varying factors including 3rd-order cycles of eustasy, and tectonic movement and sediment influx with 2nd-order changes. Instead of Exxon group's systems tracts, this model adopts parasequence sets as the fundamental building blocks of the sequence, because they are descriptive stratigraphic units simply defined by internal stacking pattern, reflecting interactions of accommodation and sediment influx. Seven sequence types which vary in number and type of internal parasequence sets are formulated as associations of four types of accommodation development and three grades of sediment influx. In the southwestern margin of Ulleung Basin, the multi-variable sequence analysis of shelf-slope sequence shows systematic changes in stratal patterns and the numbs, of constituent parasequence sets (i.e. sequence type). These changes are interpreted to reflect temporal and spatial changes in type and rate of tectonic movement and sediment influx, as a result of back-arc opening and closing. During the back-arc opening, rapid subsidence, continuous rise of relative sea level, and high sediment influx gave rise to sequences dominantly of single progradational parasequence set. In the early stage of back-arc closing accompanied by local contractional deformation, different types of sequences contemporaneously formed depending on the spatial changes in tectonically-controlled accommodation and influx rates. During the subsequent slow back-arc subsidence, rise-dominated relative sea-level cycle was coupled with moderate to high sedimentation rate to have resulted in sequences consisting of $2~3$ parasequence sets.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.69-80
• /
• 2021

Neural machine translation using deep neural network has emerged as a mainstream research, and an abundance of investment and studies on model structure and parallel language pair have been actively undertaken for the best performance. However, most recent neural machine translation studies pass along decoding strategy to future work, and have insufficient a variety of experiments and specific analysis on it for generating language to maximize quality in the decoding process. In machine translation, decoding strategies optimize navigation paths in the process of generating translation sentences and performance improvement is possible without model modifications or data expansion. This paper compares and analyzes the significant effects of the decoding strategy from classical greedy decoding to the latest Dynamic Beam Allocation (DBA) in neural machine translation using a sequence to sequence model.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.211-214
• /
• 2020

Abstract Meaning Representation(AMR)은 문장의 의미를 그래프 구조로 인코딩하여 표현하는 의미 형식표현으로 문장의 각 노드는 사건이나 개체를 취급하는 개념으로 취급하며 간선들은 이러한 개념들의 관계를 표현한다. AMR 파싱은 주어진 문장으로부터 AMR 그래프를 생성하는 자연어 처리 태스크이다. AMR 그래프의 각 개념은 추상 표현으로 문장 내의 토큰과 명시적으로 정렬되지 않는 어려움이 존재한다. 이러한 문제를 해결하기 위해 별도의 사전 학습된 정렬기를 이용하여 해결하거나 별도의 정렬기 없이 Sequence-to-Sequence 계열의 모델로 입력 문장으로부터 그래프의 노드를 생성하는 방식으로 연구되어 왔다. 본 논문에서는 문장의 입력 시퀀스와 부분 생성 그래프 사이에서 반복 추론을 통해 새로운 노드와 기존 노드와의 관계를 구성하여 점진적으로 그래프를 구성하는 모델을 한국어 AMR 데이터 셋에 적용하여 Smatch 점수 39.8%의 실험 결과를 얻었다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.157-161
• /
• 2020

문서 요약(text summarization)은 주어진 문서로부터 중요하고 핵심적인 정보를 포함하는 요약문을 만들어 내는 작업으로, 기계 번역 작업에서 주로 사용되는 Sequence-to-Sequence 모델을 사용한 end-to-end 방식의 생성(abstractive) 요약 모델 연구가 활발히 진행되고 있다. 최근에는 BERT와 MASS 같은 대용량 단일 언어 데이터 기반 사전학습(pre-training) 모델을 이용하여 미세조정(fine-tuning)하는 전이 학습(transfer learning) 방법이 자연어 처리 분야에서 주로 연구되고 있다. 본 논문에서는 MASS 모델에 복사 메커니즘(copying mechanism) 방법을 적용하고, 한국어 언어 생성(language generation)을 위한 사전학습을 수행한 후, 이를 한국어 문서 요약에 적용하였다. 실험 결과, MASS 모델에 복사 메커니즘 방법을 적용한 한국어 문서 요약 모델이 기존 모델들보다 높은 성능을 보였다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 1998.11a
• /
• pp.181-185
• /
• 1998

Recently, research and developmental direction of communication system is concurrent adopting voice data and face image in speaking to provide more higher recognition rate then in the case of only voice data. Therefore, we present a method of lipreading in speech image sequence by using the 3-D facial shape model. The method use a feature information of the face image such as the opening-level of lip, the movement of jaw, and the projection height of lip. At first, we adjust the 3-D face model to speeching face image sequence. Then, to get a feature information we compute variance quantity from adjusted 3-D shape model of image sequence and use the variance quality of the adjusted 3-D model as recognition parameters. We use the intensity inclination values which obtaining from the variance in 3-D feature points as the separation of recognition units from the sequential image. After then, we use discrete HMM algorithm at recognition process, depending on multiple observation sequence which considers the variance of 3-D feature point fully. As a result of recognition experiment with the 8 Korean vowels and 2 Korean consonants, we have about 80% of recognition rate for the plosives and vowels.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.46 no.4
• /
• pp.28-36
• /
• 2009

Sequence-Pair(SP) model represents the topological relation between modules. In general, SP model based floorplanners search solutions using Simulated-Annealing(SA) algorithm. Several SA based floorplanning techniques using SP model have been published. To improve the performance of those techniques they tried to improve the speed for evaluation function for SP model, to find better scheduling methods and perturb functions for SA. In this paper we propose a two phase SA based algorithm. In the first phase, white space between modules is reduced by applying compaction technique to the floorplan obtained by an SP. From the compacted floorplan, the corresponding SP is determined. Solution space has been searched by changing the SP in the SA framework. When solutions converge to some threshold value, the first phase of the SA based search stops. Then using the typical SA based algorithm, ie, without using the compaction technique, the second phase of our algorithm continues to find optimal solutions. Experimental results with MCNC benchmark circuits show that how the proposed technique affects to the procedure for SA based floorplainning algorithm and that the results obtained by our technique is better than those obtained by existing SA-based algorithms.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.7 no.5
• /
• pp.53-63
• /
• 2008

In this paper, we propose the regional travel time measurement algorithm using the sequence pattern matching to the type of vehicles between the origin of the region and the end of the region, that could be able to overcome the limit of conventional method such as Probe Car Method or AVI Method by License Plate Recognition. This algorithm recognizes the vehicles as a sequence group with a definite length, and measures the regional travel time by searching the sequence of the origin which is the most highly similar to the sequence of the end. According to the assumption of similarity cost function, there are proposed three types of algorithm, and it will be able to estimate the average travel time that is the most adequate to the information providing period by eliminating the abnormal value caused by inflow and outflow of vehicles. In the result of computer simulation by the length of region, the number of passing cars, the length of sequence, and the average maximum error rate are measured within 3.46%, which means that this algorithm is verified for its superior performance.

• Korean Journal of Cognitive Science
• /
• v.18 no.2
• /
• pp.179-191
• /
• 2007

In a goal-oriented dialogue, spoken' intentions can be represented by domain actions that consist of pairs of a speech art and a concept sequence. The domain action prediction of user's utterance is useful to correct some errors that occur in a speech recognition process, and the domain action prediction of system's utterance is useful to generate flexible responses. In this paper, we propose a model to predict a domain action of the next utterance using a neural network. The proposed model predicts the next domain action by using a dialogue history vector and a current domain action as inputs of the neural network. In the experiment, the proposed model showed the precision of 80.02% in speech act prediction and the precision of 82.09% in concept sequence prediction.

• Journal of KIISE:Software and Applications
• /
• v.35 no.9
• /
• pp.554-561
• /
• 2008

Prediction technique of user's intention can be used as a post-processing method for reducing the search space of an automatic speech recognizer. Prediction technique of system's intention can be used as a pre-processing method for generating a flexible sentence. To satisfy these practical needs, we propose a statistical model to predict speakers' intentions that are generalized into pairs of a speech act and a concept sequence. Contrary to the previous model using simple n-gram statistic of speech acts, the proposed model represents a dialogue history of a current utterance to a feature set with various linguistic levels (i.e. n-grams of speech act and a concept sequence pairs, clue words, and state information of a domain frame). Then, the proposed model predicts the intention of the next utterance by using the feature set as inputs of CRFs (Conditional Random Fields). In the experiment in a schedule management domain, The proposed model showed the precision of 76.25% on prediction of user's speech act and the precision of 64.21% on prediction of user's concept sequence. The proposed model also showed the precision of 88.11% on prediction of system's speech act and the Precision of 87.19% on prediction of system's concept sequence. In addition, the proposed model showed 29.32% higher average precision than the previous model.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.34-36
• /
• 2019

기존 딥 러닝을 이용한 형태소 분석 및 품사 태깅(Part-Of-Speech tagging)은 feed-forward neural network에 CRF를 결합하는 방법이나 sequence-to-sequence 모델을 이용한 방법 등의 다양한 모델들이 연구되었다. 본 논문에서는 한국어 형태소 분석 및 품사 태깅을 수행하기 위하여 최근 자연어처리 태스크에서 많은 성능 향상을 보이고 있는 BERT를 기반으로 한 음절 단위 LSTM-CRF 모델을 제안한다. BERT는 양방향성을 가진 트랜스포머(transformer) 인코더를 기반으로 언어 모델을 사전 학습한 것이며, 본 논문에서는 한국어 대용량 코퍼스를 어절 단위로 사전 학습한 KorBERT를 사용한다. 실험 결과, 본 논문에서 제안한 모델이 기존 한국어 형태소 분석 및 품사 태깅 연구들 보다 좋은 (세종 코퍼스) F1 98.74%의 성능을 보였다.