• Communications for Statistical Applications and Methods
• /
• 제30권5호
• /
• pp.501-516
• /
• 2023

Knowledge distillation (KD) refers to extracting knowledge from a large and complex model (teacher) and transferring it to a relatively small model (student). This can be done by training the teacher model to obtain the activation function values of the hidden or the output layers and then retraining the student model using the same training data with the obtained values. Recently, relational KD (RKD) has been proposed to extract knowledge about relative differences in training data. This method improved the performance of the student model compared to conventional KDs. In this paper, we propose a new method for RKD by introducing a new loss function for RKD. The proposed loss function is defined using the area difference between the teacher model and the student model in a specific hidden layer, and it is shown that the model can be successfully compressed, and the generalization performance of the model can be improved. We demonstrate that the accuracy of the model applying the method proposed in the study of model compression of audio data is up to 1.8% higher than that of the existing method. For the study of model generalization, we demonstrate that the model has up to 0.5% better performance in accuracy when introducing the RKD method to self-KD using image data.

• 정보처리학회논문지B
• /
• 제14B권3호
• /
• pp.231-240
• /
• 2007

이 논문은 정렬 기법을 이용한 미등록 대역어 추출 모델을 제안하고 그 추출 시스템을 구현한다. 제안된 미등록 대역어 추출 모델은 일종의 구절정렬 모델로서 경계모델과 언어모델 그리고 번역 모델로 구성된다. 제안된 추출 시스템은 병렬말뭉치 구축, 단어정렬, 미등록어 추출로 구성된다. 이 논문에서는 제안된 시스템을 평가하기 위해서 약 1,500여 개의 미등록어가 포함된 2,200문장의 평가말뭉치를 구축하여 다양한 실험을 수행하였다. 실험을 통해서 제안된 모델이 미등록 대역어 추출에 매우 유용함을 알 수 있었다. 앞으로 좀 더 객관적인 평가를 위해 대량의 평가말뭉치 구축이 선행되어야 하며 좀 더 양질의 병렬말뭉치의 구축이 필요할 것이다. 또한 미등록어 추출 모델을 개선하기 다양한 연구가 추진되어야 할 것이다.

• 한국전산구조공학회:학술대회논문집
• /
• 한국전산구조공학회 1989년도 봄 학술발표회 논문집
• /
• pp.50-55
• /
• 1989

Piping systems are composed of pipes with various thickness, diameter and length. Accurate analysis of a piping system requires a complicated three dimensional finite element model and a computer system with large memory size, while simplified models result in system response prediction with deteriorated accuracy. An efficient analysis model for piping systems is proposed in this study. The proposed model is developed by introducing a joint model which accounts for the behavior of a pipe connection. Pipes are represented by beam elements and the effect of local deformation of pipe connections are replaced by joint element deformations. The proposed model which is as simple and efficient as a beam model can be used to obtain piping system response with accuracy close to that of a finite element model.

• 수산경영론집
• /
• 제47권4호
• /
• pp.1-13
• /
• 2016

Fish mortality is the most important success factor in aquaculture management. To analyze the effect of mortality considering biological and economic condition is a important problem in land-based aquaculture. This study is aimed to analyze the effect of mortality for duration of cultivation in land-based aquaculture. This study builds the mathematical model that finds the value of decision variable to minimize cost that sums up the water pool usage cost, sorting cost, fingerling cost and feeding cost under critical standing corp constraint. The proposed mathematical model involves many aspects, both biological and economical: (1) number of fingerlings (2) timing and number of batch splitting event, based on (3) fish growth rate, (4) mortality, and (5) several farming expense. Numerical simulation model presented here in. The objective of numerical simulation is to provide for decision makers to analyse and comprehend the proposed model. When extensive biological and cost data become available, the proposed model can be widely applied to yield more accurate results.

• 한국군사과학기술학회지
• /
• 제14권5호
• /
• pp.818-824
• /
• 2011

In this paper, we proposed a phase noise model of a frequency synthesizer for a radar system. Especially, it was proposed a phase noise model in a DAS(Direct Analog Synthesizer) and a frequency up converter system using Leeson's model. The proposed phase noise model was derived from the measurement data of model 1 and evaluated by adapting to model 2 and model 3 frequency synthesizers. The prediction phase noise by modeling was totally matched to the measured data and the effective analysis of the phase noise was done in a frequency synthesizer and a frequency converter of radar system.

• Journal of Electrical Engineering and Technology
• /
• 제8권5호
• /
• pp.1243-1250
• /
• 2013

A model of a permanent magnet synchronous motor (PMSM) with an inter turn short fault is proposed using a deformed flux model. The deformed flux model includes not only the fault winding flux information but also the inductance variation of the healthy winding considering the configuration of the winding distribution. With the deformed flux model and the positive sequence current assumption, the proposed model is derived in the positive and negative sequence synchronous reference frame (SRF). The finite elements method (FEM) simulation is applied to validate the proposed PMSM model with inter turn short fault.

• 조명전기설비학회논문지
• /
• 제27권6호
• /
• pp.59-67
• /
• 2013

This paper presents a neural net type process model for enhancing learning compensation function in hot strip finishing rolling mill. Adequate input and output variables of process model are chosen, the proposed model was designed as single layer neural net. Equivalent carbon content, strip thickness and rolling speed are suggested as input variables, and looper's manipulation variable is proposed as output variable. According to simulation result using process data to show the validity of the proposed process model, neural net type process model's outputs give almost similar data to process output under same input conditions.

• Steel and Composite Structures
• /
• 제17권4호
• /
• pp.371-386
• /
• 2014

This paper presents a tri-linear restoring force model based on the test results of 12 circular RC columns strengthened by CFRP strips under low cyclic loading. The pre-stress of CFRP strips and axial load ratio of specimens are considered as the affect parameters of the proposed model. All essential characteristics of the hysteretic behavior of the proposed model, including the hysteretic rules, main performance points, strength degradation, stiffness degradation and confinement effects are explicitly analyzed. The calculated results from the proposed model are in good agreement with the experimental results, which shows that the recommended model can be reliably used for seismic behavior predictions of circular RC columns strengthened by pre-stressed CFRP strips.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2001년도 춘계학술대회 논문집
• /
• pp.92-97
• /
• 2001

The major faults of CNC machine tool is operational error which is charge over 70%. This paper describes model of remote service and fault diagnosis for CNC machine tool with open architecture controller. For intelligent fault diagnosis, new model is proposed. In this paper, the three major operational faults, emergency stop error, cycle start disable and machine ready disable, are defined. Two diagnostic models based on the ladder diagram, switching function model, step switching function model, are proposed. For internet based remote service, suitable environment is proposed and implemented with web server and client.

• ETRI Journal
• /
• 제38권3호
• /
• pp.487-493
• /
• 2016

Two new methods are proposed for an unsupervised adaptation of a language model (LM) with a single sentence for automatic transcription tasks. At the training phase, training documents are clustered by a method known as Latent Dirichlet allocation (LDA), and then a domain-specific LM is trained for each cluster. At the test phase, an adapted LM is presented as a linear mixture of the now trained domain-specific LMs. Unlike previous adaptation methods, the proposed methods fully utilize a trained LDA model for the estimation of weight values, which are then to be assigned to the now trained domain-specific LMs; therefore, the clustering and weight-estimation algorithms of the trained LDA model are reliable. For the continuous speech recognition benchmark tests, the proposed methods outperform other unsupervised LM adaptation methods based on latent semantic analysis, non-negative matrix factorization, and LDA with n-gram counting.