• International journal of advanced smart convergence
• /
• 제8권2호
• /
• pp.132-139
• /
• 2019

We address the problem about forecasting the direction of stock price movement in the Korea market. Recently, the deep neural network is popularly applied in this area of research. In deep neural network systems, proper parameter initialization reduces training time and improves the performance of the model. Therefore, in our study, we propose a novel parameter initialization technique and apply this technique for the stock price movement prediction model. Specifically, we design a framework which consists of two models: a base model and a main prediction model. The base model constructed with LSTM is trained by using the large data which is generated by a large amount of the stock data to achieve optimal parameters. The main prediction model with the same architecture as the base model uses the optimal parameter initialization. Thus, the main prediction model is trained by only using the data of the given stock. Moreover, the stock price movements can be affected by other related information in the stock market. For this reason, we conducted our research with two types of inputs. The first type is the stock features, and the second type is a combination of the stock features and the Korea Composite Stock Price Index (KOSPI) features. Empirical results conducted on the top five stocks in the KOSPI list in terms of market capitalization indicate that our approaches achieve better predictive accuracy and F1-score comparing to other baseline models.

• 한국멀티미디어학회논문지
• /
• 제21권4호
• /
• pp.441-449
• /
• 2018

Deep learning has been used for various processing centered on image recognition. One core algorithms of the deep learning, convolutional neural network is an deep neural network that specialized in image recognition. In this paper, we use a convolutional neural network to classify forest insects and propose an optimization method. Experiments were carried out by combining two weight initialization and six parameter update methods. As a result, the Xavier-SGD method showed the highest performance with an accuracy of 82.53% in the 12 different combinations of experiments. Through this, the latest learning algorithms, which complement the disadvantages of the previous parameter update method, we conclude that it can not lead to higher performance than existing methods in all application environments.

• Journal of Electrical Engineering and Technology
• /
• 제11권6호
• /
• pp.1846-1856
• /
• 2016

This study proposes a novel feature point initialization method in order to improve the accuracy of feature point positions by fusing a vision sensor and a lidar. The initialization is a process that determines three dimensional positions of feature points through two dimensional image data, which has a direct influence on performance of a 6-DoF bearing-only SLAM. Prior to the initialization, an extrinsic calibration method which estimates rotational and translational relationships between a vision sensor and lidar using multiple calibration tools was employed, then the feature point initialization method based on the estimated extrinsic calibration parameters was presented. In this process, in order to improve performance of the accuracy of the initialized feature points, an iterative automatic scaling parameter tuning technique was presented. The validity of the proposed feature point initialization method was verified in a 6-DoF bearing-only SLAM framework through an indoor and outdoor tests that compare estimation performance with the previous initialization method.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2003년도 ICCAS
• /
• pp.752-757
• /
• 2003

Recently, 3D structure recovery through self-calibration of camera has been actively researched. Traditional calibration algorithm requires known 3D coordinates of the control points while self-calibration only requires the corresponding points of images, thus it has more flexibility in real application. In general, self-calibration algorithm results in the nonlinear optimization problem using constraints from the intrinsic parameters of the camera. Thus, it requires initial value for the nonlinear minimization. Traditional approaches get the initial values assuming they have the same intrinsic parameters while they are dealing with the situation where the intrinsic parameters of the camera may change. In this paper, we propose new initialization method using the minimum 2 images. Proposed method is based on the assumption that the least violation of the camera’s intrinsic parameter gives more stable initial value. Synthetic and real experiment shows this result.

• 음성과학
• /
• 제5권1호
• /
• pp.7-21
• /
• 1999

This paper proposes an improved training procedure in speech recognition based on the continuous density of the Hidden Markov Model (CDHMM). Of the three parameters (initial state distribution probability, state transition probability, output probability density function (p.d.f.) of state) governing the CDHMM model, we focus on the third parameter and propose an efficient algorithm that determines the p.d.f. of each state. It is known that the resulting CDHMM model converges to a local maximum point of parameter estimation via the iterative Expectation Maximization procedure. Specifically, we propose two independent algorithms that can be embedded in the segmental K -means training procedure by replacing relevant key steps; the adaptation of the number of mixture Gaussian p.d.f. and the initialization using the CDHMM parameters previously estimated. The proposed adaptation algorithm searches for the optimal number of mixture Gaussian humps to ensure that the p.d.f. is consistently re-estimated, enabling the model to converge toward the global maximum point. By applying an appropriate threshold value, which measures the amount of collective changes of weighted variances, the optimized number of mixture Gaussian branch is determined. The initialization algorithm essentially exploits the CDHMM parameters previously estimated and uses them as the basis for the current initial segmentation subroutine. It captures the trend of previous training history whereas the uniform segmentation decimates it. The recognition performance of the proposed adaptation procedures along with the suggested initialization is verified to be always better than that of existing training procedure using fixed number of mixture Gaussian p.d.f.

• 대한기계학회논문집
• /
• 제8권1호
• /
• pp.65-70
• /
• 1984

A recursive parameter estimation method is applied to spindle deflection model during boring process. The spindle infeed rate is then determined to preserve the diametral tolerance of bore. This estimation method is further extended to adaptive control by application of the variance perturbation method. The results of computer simulation attest that the proposed method renders the optimal cutting conditions, maintaining the diametral accuracy of bore, regardless of parameter fluctuations. The proposed method necessitating only post-process measurements features that initialization of parameter guess values in simple, a priori knowledge on parameter variations is not needed and the accurate estimation of optimal spindle infeed rate is obtained, even if the parameter estimation may be poor.

• 전기전자학회논문지
• /
• 제22권1호
• /
• pp.14-20
• /
• 2018

The purpose of training a convolutional neural network (CNN) is to obtain weight factors that give high classification accuracies. The initial values of hyper-parameters affect the training results, and it is important to train a CNN with a suitable hyper-parameter set of a learning rate, a batch size, the initialization of weight factors, and an optimizer. We investigate the effects of a single hyper-parameter while others are fixed in order to obtain a hyper-parameter set that gives higher classification accuracies and requires shorter training time using a proposed VGG-like CNN for training since the VGG is widely used. The CNN is trained for four datasets of CIFAR10, CIFAR100, GTSRB and DSDL-DB. The effects of the normalization and the data transformation for datasets are also investigated, and a training scheme using merged datasets is proposed.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2001년도 ICCAS
• /
• pp.169.4-169
• /
• 2001

Conventional methods to solve the nonlinear programming problem range from augmented Lagrangian methods to sequential quadratic programming (SQP) methods. NPSOL, which is a SQP code, has been widely used to solve various optimization problems but is still subject to many numerical problems such as convergence to local optima, difficulties in initialization and in handling non-smooth cost functions. Recently, many evolutionary methods have been developed for constrained optimization. Among them, CEALM (Co-Evolutionary Augmented Lagrangian Method) shows excellent performance in the following aspects: global optimization capability, low sensitivity to the initial parameter guessing, and excellent constraint handling capability due to the benefit of the augmented Lagrangian function. This algorithm is ...

• Journal of Microbiology and Biotechnology
• /
• 제14권6호
• /
• pp.1163-1169
• /
• 2004

An unstructured mathematical model is presented for lactic acid fermentation based on the energy balance. The proposed model reflects the energy metabolic state and then predicts the cell growth, lactic acid production, and glucose consumption rates by relating the above rates with the energy metabolic rate. Fermentation experiments were conducted under various initial lactic acid concentrations of 0, 30, 50, 70, and 90 g/l. Also, a genetic algorithm was used for further optimization of the model parameters and included the operations of coding, initialization, hybridization, mutation, decoding, fitness calculation, selection, and reproduction exerted on individuals (or chromosomes) in a population. The simulation results showed a good fit between the model prediction and the experimental data. The genetic algorithm proved to be useful for model parameter optimization, suggesting wider applications in the field of biological engineering.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 1999년도 KOBA 방송기술 워크샵 KOBA Broadcasting Technology Workshop
• /
• pp.125-131
• /
• 1999

In this paper, we address practical methods for calculating camera lens distortion for real time applications. Although the lens distortion problem can be easily ignored for constant-parameter lenses, in the field of real-time camera calibrations, for zoom lenses a large number of calculations are needed to calculate the distortion. However, if the distortion can be calculated independently of the other camera parameter, we can easily calibrate a camera without the need for a large number of calculations. Based on Tsai's camera model, we propose two different methods for calculating lens distortion. These methods are so simple and require so few calculations that the lens distortion can be rapidly calculated even in real-time applications. The first method is to refer to the focal length - lens distortion Look Up Table(LUT), which is constructed in the initialization process. The second method is to use the relationship between the feature points found in the image. Experiments were carried out for both methods, results of which show that the proposed methods are favorably comparable in performance with non-real full optimization method.