• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.6
• /
• pp.347-353
• /
• 2022

In this paper, we propose a method that performs a text-video retrieval model by replacing video properties using captions. In general, the exisiting embedding-based models consist of both joint embedding space construction and the CNN-based video encoding process, which requires a lot of computation in the training as well as the inference process. To overcome this problem, we introduce a video-captioning module to replace the visual property of video with captions generated by the video-captioning module. To be specific, we adopt the caption generator that converts candidate videos into captions in the inference process, thereby enabling direct comparison between the text given as a query and candidate videos without joint embedding space. Through the experiment, the proposed model successfully reduces the amount of computation and inference time by skipping the visual processing process and joint embedding space construction on two benchmark dataset, MSR-VTT and VATEX.

• Proceedings of the Korea Contents Association Conference
• /
• 2009.05a
• /
• pp.105-111
• /
• 2009

Widely used color information recognition methods based on the RGB color model with static fuzzy inference rules have limitations due to the model itself - the detachment of human vision and applicability of limited environment. In this paper, we propose a method that is based on HSI model with new inference process that resembles human vision recognition process. Also, a user can add, delete, update the inference rules in this system. In our method, we design membership intervals with sine, cosine function in H channel and with functions in trigonometric style in S and I channel. The membership degree is computed via interval merging process. Then, the inference rules are applied to the result in order to infer the color information. Our method is proven to be more intuitive and efficient compared with RGB model in experiment.

• Journal of the Korean Data and Information Science Society
• /
• v.22 no.6
• /
• pp.1017-1028
• /
• 2011

When it is not easy to decide the credit scoring for some loan applicants, credit evaluation is postponded and reserve to ask a specialist for further evaluation of undecided applicants. This undecided inference is one of problems that happen to most statistical models including the biostatistics and sportal statistics as well as credit evaluation area. In this work, the undecided inference is regarded as a missing data mechanism under the assumption of MNAR, and use the bivariate probit model which is one of sample selection models. Two undecided inference methods are proposed: one is to make use of characteristic variables to represent the state for decided applicants, and the other is that more accurate and additional informations are collected and apply these new variables. With an illustrated example, misclassification error rates for undecided and overall applicants are obtainded and compared according to various characteristic variables, undecided intervals, and thresholds. It is found that misclassification error rates could be reduced when the undecided interval is increased and more accurate information is put to model, since more accurate situation of decided applications are reflected in the bivariate probit model.

• Phonetics and Speech Sciences
• /
• v.13 no.4
• /
• pp.47-53
• /
• 2021

Autoregressive Text-to-Speech (TTS) models suffer from inference instability and slow inference speed. Inference instability occurs when a poorly predicted sample at time step t affects all the subsequent predictions. Slow inference speed arises from a model structure that forces the predicted samples from time steps 1 to t-1 to predict the sample at time step t. In this study, an end-to-end non-autoregressive fast text-to-speech model is suggested as a solution to these problems. The results of this study show that this model's Mean Opinion Score (MOS) is close to that of Tacotron 2 - WaveNet, while this model's inference speed and stability are higher than those of Tacotron 2 - WaveNet. Further, this study aims to offer insight into the improvement of non-autoregressive models.

• Smart Structures and Systems
• /
• v.15 no.4
• /
• pp.1139-1158
• /
• 2015

This paper proposes an efficient system identification method for modeling nonlinear behavior of civil structures. This method is developed by integrating three different methodologies: principal component analysis (PCA), artificial neural networks, and fuzzy logic theory, hence named PANFIS (PCA-based adaptive neuro-fuzzy inference system). To evaluate this model, a 3-story building equipped with a magnetorheological (MR) damper subjected to a variety of earthquakes is investigated. To train the input-output function of the PANFIS model, an artificial earthquake is generated that contains a variety of characteristics of recorded earthquakes. The trained model is also validated using the1940 El-Centro, Kobe, Northridge, and Hachinohe earthquakes. The adaptive neuro-fuzzy inference system (ANFIS) is used as a baseline. It is demonstrated from the training and validation processes that the proposed PANFIS model is effective in modeling complex behavior of the smart building. It is also shown that the proposed PANFIS produces similar performance with the benchmark ANFIS model with significant reduction of computational loads.

• ETRI Journal
• /
• v.41 no.6
• /
• pp.771-781
• /
• 2019

We propose the VS³-NET model to solve the task of question answering questions with machine-reading comprehension that searches for an appropriate answer in a given context. VS³-NET is a model that trains latent variables for each question using variational inferences based on a model of a simple recurrent unit-based sentences and self-matching networks. The types of questions vary, and the answers depend on the type of question. To perform efficient inference and learning, we introduce neural question-type models to approximate the prior and posterior distributions of the latent variables, and we use these approximated distributions to optimize a reparameterized variational lower bound. The context given in machine-reading comprehension usually comprises several sentences, leading to performance degradation caused by context length. Therefore, we model a hierarchical structure using sentence encoding, in which as the context becomes longer, the performance degrades. Experimental results show that the proposed VS³-NET model has an exact-match score of 76.8% and an F1 score of 84.5% on the SQuAD test set.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.11 no.5
• /
• pp.395-399
• /
• 2001

This paper describes modular fuzzy inference systems(MFIS) with adaptive capability to extract fuzzy inference modules from observation data through the learning process. The proposed MFIS is based on the structural similarity to Tagaki-Sugeno fuzzy models and a modular neural architecture. The learning of MFIS is done by assigning new fuzzy inference modules and by updating the parameters of existing modules. The fuzzy inference modules consist of local model network and fuzzy gating network. The parameters of the MFIS are updated by the standard LMS algorithm. The performance of the MFIS is illustrated with adaptive control of a nonlinear dynamic system.

• KDI Journal of Economic Policy
• /
• v.38 no.2
• /
• pp.1-20
• /
• 2016

In this paper, we conduct uniform inference of two widely used versions of the Phillips curve, specifically the random-walk Phillips curve and the New-Keynesian Phillips curve (NKPC). For both specifications, we propose a potentially time-varying natural unemployment (NAIRU) to address the uncertainty surrounding the inflation-unemployment trade-off. The inference is conducted through the construction of what is known as the uniform confidence band (UCB). The proposed methodology is then applied to point-ahead inflation forecasting for the Korean economy. This paper finds that the forecasts can benefit from conducting UCB-based inference and that the inference results have important policy implications.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.52 no.5
• /
• pp.283-289
• /
• 2003

This paper proposes an expert system which can determine automatically the shunting routes corresponding to the given shunting works by considering totally the train operating environments in the station. The expert system proposes the multiple shunting routes with priority of selection based on heuristic search strategy. Accordingly, system operator can select a shunting route with the safety and efficiency among the those shunting routes. The expert system consists of a main inference engine and a sub inference engine. The main inference engine determines the shunting routes with selection priority using the segment routes obtained from the sub inference engine. The heuristic rules are extracted from operating knowledges of the veteran route operator and station topology. It is implemented in C computer language for the purpose of the implementation of the inference engine using the dynamic memory allocation technique. And, the validity of the builted expert system is proved by a test case for the model station.

• Asian Pacific Journal of Cancer Prevention
• /
• v.15 no.2
• /
• pp.663-669
• /
• 2014

Background: With recent progress in health science administration, a huge amount of data has been collected from thousands of subjects. Statistical and computational techniques are very necessary to understand such data and to make valid scientific conclusions. The purpose of this paper was to develop a statistical probability model and to predict future survival times for male breast cancer patients who were diagnosed in the USA during 1973-2009. Materials and Methods: A random sample of 500 male patients was selected from the Surveillance Epidemiology and End Results (SEER) database. The survival times for the male patients were used to derive the statistical probability model. To measure the goodness of fit tests, the model building criterions: Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), and Deviance Information Criteria (DIC) were employed. A novel Bayesian method was used to derive the posterior density function for the parameters and the predictive inference for future survival times from the exponentiated Weibull model, assuming that the observed breast cancer survival data follow such type of model. The Markov chain Monte Carlo method was used to determine the inference for the parameters. Results: The summary results of certain demographic and socio-economic variables are reported. It was found that the exponentiated Weibull model fits the male survival data. Statistical inferences of the posterior parameters are presented. Mean predictive survival times, 95% predictive intervals, predictive skewness and kurtosis were obtained. Conclusions: The findings will hopefully be useful in treatment planning, healthcare resource allocation, and may motivate future research on breast cancer related survival issues.