• 한국통신학회논문지
• /
• 제30권6C호
• /
• pp.481-489
• /
• 2005

In this letter, we propose an efficient training sequence structure for adaptive linear multiuser detectors in space-time block coded multiuser systems, by exploiting a particular property of the minimum mean square error multiuser detectors used in these systems. The proposed structure wastes less overall system capacity than the straightforward training structure, without any corresponding loss of performance, as confirmed by the simulation results.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제14권3호
• /
• pp.209-215
• /
• 2014

Sequence tagging is the task of predicting frame-wise labels for a given input sequence and has important applications to diverse domains. Conventional methods such as maximum likelihood (ML) learning matches global features in empirical and model distributions, rather than local features, which directly translates into frame-wise prediction errors. Recent probabilistic sequence models such as conditional random fields (CRFs) have achieved great success in a variety of situations. In this paper, we introduce a novel discriminative CRF learning algorithm to minimize local feature mismatches. Unlike overall data fitting originating from global feature matching in ML learning, our approach reduces the total error over all frames in a sequence. We also provide an efficient gradient-based learning method via gradient forward-backward recursion, which requires the same computational complexity as ML learning. For several real-world sequence tagging problems, we empirically demonstrate that the proposed learning algorithm achieves significantly more accurate prediction performance than standard estimators.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1989년도 한국자동제어학술회의논문집; Seoul, Korea; 27-28 Oct. 1989
• /
• pp.772-775
• /
• 1989

It is difficult to obtain a swinging-up control sequence of a one-link pendulum analytically or numerically. In this paper, we obtain a proper control sequence through manual control experiments. However, no proper control sequence will be obtained if the rotational velocity of the pendulum is fast for the human operator. To overcome such a disadvantage, we propose a method for training the operator by using a pendulum simulator.

• 한국잠사곤충학회지
• /
• 제40권1호
• /
• pp.33-37
• /
• 1998

Two types of sperm, apyrene and eupyrene, are identified in A. mylitta. The sperm in the adult moth are motionless in seminal vesicles. At the time of ejaculation they received a secretion from male ejaculatory duct that renders them motile. The dissociation of eupyrene bundles, eupyrene sperm motility and the sequence of events of sperm migration in both sexes are described in the present paper.

• 한국전자파학회논문지
• /
• 제10권7호
• /
• pp.1085-1094
• /
• 1999

유럽의 GSM에서는 시간 분할 다중 접속(TDMA) 프레임에 정보 선호와 함께 한 프레임당 페이딩 채널의 임펄스 응답 측정과 학습 구간 동안의 등화기의 빠른 수렴을 목적으로 한 16비트 길이의 프리앵블을 사용하고 있다. 이러한 전송 프리엠블과 수선기에 저장된 프리엠블에 일정 조건을 만족하는 동일한 이진 부호를 전송하여 등화기의 랩 계수(tap coefficients)를 빠르게 수렴시키고, 그 상호 상관 함수를 구함으로써 임펄스 응답을 측정한다. 본 논문에서는 기존의 의사 잡음 부호를 이용하는 방식과는 달리 0번 자리 이동(shift)일 때와 반 주기만큼의 자리 이동일 때 상관 값을 가지고, 나머지 자리 이동에서는 O값인 특정 자기 상관 함수를 가지는 이진 부호를 이용한 정확한 임펄스 응답 측정과 초기 빠른 수렴 속도를 지니는 등화기(equalizer) 구현의 수학적 접근과 기존의 의사 잡음 부호를 학습 시권스로 이용한 경우와의 수렴 속도 비교, 그리고 임펄 스 응답 측정이 논의된다.

• International Journal of Internet, Broadcasting and Communication
• /
• 제12권2호
• /
• pp.45-50
• /
• 2020

A Language model with neural networks commonly trained with likelihood loss. Such that the model can learn the sequence of human text. State-of-the-art results achieved in various language generation tasks, e.g., text summarization, dialogue response generation, and text generation, by utilizing the language model's next token output probabilities. Monotonous and boring outputs are a well-known problem of this model, yet only a few solutions proposed to address this problem. Several decoding techniques proposed to suppress repetitive tokens. Unlikelihood training approached this problem by penalizing candidate tokens probabilities if the tokens already seen in previous steps. While the method successfully showed a less repetitive generated token, the method has a large memory consumption because of the training need a big vocabulary size. We effectively reduced memory footprint by encoding words as sequences of subword units. Finally, we report competitive results with token level unlikelihood training in several automatic evaluations compared to the previous work.

• 한국멀티미디어학회논문지
• /
• 제19권1호
• /
• pp.22-29
• /
• 2016

Hidden Markov Model is one of the most successful and popular tools for modeling real world sequential data. Real world signals come in a variety of shapes and variabilities, among which temporal and spectral ones are the prime targets that the HMM aims at. A new problem that is gaining increasing attention is characterizing missing observations in incomplete data sequences. They are incomplete in that there are holes or omitted measurements. The standard HMM algorithms have been developed for complete data with a measurements at each regular point in time. This paper presents a modified algorithm for a discrete HMM that allows substantial amount of omissions in the input sequence. Basically it is a variant of Baum-Welch which explicitly considers the case of isolated or a number of omissions in succession. The algorithm has been tested on online handwriting samples expressed in direction codes. An extensive set of experiments show that the HMM so modeled are highly flexible showing a consistent and robust performance regardless of the amount of omissions.

• 한국정보통신학회:학술대회논문집
• /
• 한국해양정보통신학회 2006년도 춘계종합학술대회
• /
• pp.378-381
• /
• 2006

본 논문은 ISI가 존재하는 MIMO-OFDM 시스템에 사용되는 새로운 채널 추정 방법을 제안한다. 제안된 방법은 OFDM에서 문제가 되는 PAR를 일정하게 하고, IST의 영향을 제거하는 특별한 학습 신호를 사용한다. 이러한 학습 계열을 사용함으로서, LS(Least Square) 추정에서 문제가 되는 행렬의 특이를 피할 수 있다. 시뮬레이션 결과, SNR이 클 때 제안된 방법은 전통적인 방법보다 추정된 채널의 MSE가 20dB 이상 우수함을 확인할 수 있다.

• The Journal of the Acoustical Society of Korea
• /
• 제15권4E호
• /
• pp.21-27
• /
• 1996

In this paper, we propose a discriminative training algorithm for the stochastic segment model (SSM) in continuous speech recognition. As the SSM is usually trained by maximum likelihood estimation (MLE), a discriminative training algorithm is required to improve the recognition performance. Since the SSM does not assume the conditional independence of observation sequence as is done in hidden Markov models (HMMs), the search space for decoding an unknown input utterance is increased considerably. To reduce the computational complexity and starch space amount in an iterative training algorithm for discriminative SSMs, a hybrid architecture of SSMs and HMMs is programming using HMMs. Given the segment boundaries, the parameters of the SSM are discriminatively trained by the minimum error classification criterion based on a generalized probabilistic descent (GPD) method. With the discriminative training of the SSM, the word error rate is reduced by 17% compared with the MLE-trained SSM in speaker-independent continuous speech recognition.

• 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.