• Journal of Information Processing Systems
• /
• v.15 no.2
• /
• pp.422-432
• /
• 2019

A hybrid kernel function of support vector machine is proposed to improve the classification performance of power quality disturbances. The kernel function mathematical model of support vector machine directly affects the classification performance. Different types of kernel functions have different generalization ability and learning ability. The single kernel function cannot have better ability both in learning and generalization. To overcome this problem, we propose a hybrid kernel function that is composed of two single kernel functions to improve both the ability in generation and learning. In simulations, we respectively used the single and multiple power quality disturbances to test classification performance of support vector machine algorithm with the proposed hybrid kernel function. Compared with other support vector machine algorithms, the improved support vector machine algorithm has better performance for the classification of power quality signals with single and multiple disturbances.

• Journal of KIISE:Software and Applications
• /
• v.30 no.3_4
• /
• pp.326-338
• /
• 2003

The objective of a neural network design and model selection is to construct an optimal network with a good generalization performance. However, training data include noises, and the number of training data is not sufficient, which results in the difference between the true probability distribution and the empirical one. The difference makes the teaming parameters to over-fit only to training data and to deviate from the true distribution of data, which is called the overfitting phenomenon. The overfilled neural network shows good approximations for the training data, but gives bad predictions to untrained new data. As the complexity of the neural network increases, this overfitting phenomenon also becomes more severe. In this paper, by taking statistical viewpoint, we proposed an integrative process for neural network design and model selection method in order to improve generalization performance. At first, by using the natural gradient learning with adaptive regularization, we try to obtain optimal parameters that are not overfilled to training data with fast convergence. By adopting the natural pruning to the obtained optimal parameters, we generate several candidates of network model with different sizes. Finally, we select an optimal model among candidate models based on the Bayesian Information Criteria. Through the computer simulation on benchmark problems, we confirm the generalization and structure optimization performance of the proposed integrative process of teaming and model selection.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.7 no.7
• /
• pp.1705-1720
• /
• 2013

Human face gender recognition requires fast image processing with high accuracy. Existing face gender recognition methods used traditional local features and machine learning methods have shortcomings of low accuracy or slow speed. In this paper, a new framework for face gender recognition to reach fast face gender recognition is proposed, which is based on Local Ternary Pattern (LTP) and Extreme Learning Machine (ELM). LTP is a generalization of Local Binary Pattern (LBP) that is in the presence of monotonic illumination variations on a face image, and has high discriminative power for texture classification. It is also more discriminate and less sensitive to noise in uniform regions. On the other hand, ELM is a new learning algorithm for generalizing single hidden layer feed forward networks without tuning parameters. The main advantages of ELM are the less stringent optimization constraints, faster operations, easy implementation, and usually improved generalization performance. The experimental results on public databases show that, in comparisons with existing algorithms, the proposed method has higher precision and better generalization performance at extremely fast learning speed.

• The KIPS Transactions:PartB
• /
• v.12B no.3 s.99
• /
• pp.343-348
• /
• 2005

The goal of this paper is to study the joint effect of factors of neural network teaming procedure. There are many factors, which may affect the generalization ability and teaming speed of neural networks, such as the initial values of weights, the learning rates, and the regularization coefficients. We will apply a constructive training algerian for neural network, then patterns are trained incrementally by considering them one by one. First, we will investigate the effect of these factors on generalization performance and learning speed. Based on these factors' effect, we will propose a joint method that simultaneously considers these three factors, and dynamically hue the learning rate and regularization coefficient. Then we will present the results of some experimental comparison among these kinds of methods in several simulated nonlinear data. Finally, we will draw conclusions and make plan for future work.

• Proceedings of the IEEK Conference
• /
• 2008.06a
• /
• pp.1017-1018
• /
• 2008

In this paper, we propose a method that makes use of neighborhood relationship in 2D spectrogram of separated sources toward the generalization of the binary mask in Degenerate Unmixing Estimation Technique (DUET). A new generalized mask can be consist of five to ten mask. According to the new mask, the original power of the spectrogram in each frequency-time point is assigned. The result showed a smooth and tender wave-form, indicating a high speech separation performance compared to the original method.

• The Korean Journal of Applied Statistics
• /
• v.17 no.1
• /
• pp.75-88
• /
• 2004

The Support Vector Machine(SVM) algorithm has paid attention on maximizing the shortest distance between sample points and discrimination hyperplane. This paper suggests the total margin algorithm which considers the distance between all data points and the separating hyperplane. The method extends existing support vector machine algorithm. In addition, this newly proposed method improves the generalization error bound. Numerical experiments show that the total margin algorithm provides good performance, comparing with the previous methods.

• International Journal of Advanced Culture Technology
• /
• v.6 no.4
• /
• pp.80-86
• /
• 2018

The intervention of Autism Spectrum Disorder (ASD) is limited research focus on the effect of occupational-based intervention. This study sought to determine the effect of occupational-based intervention of chopstick skills for children with ASD. This study included a total of 3 children with ASD.Using single-subject study design, a changing criterion design and ABC design were implemented. The participants' behavior was observed and recorded throughout each session. In this study, the results were analyzed through visual graphs. The amount of food that was moved using the chopsticks was gradually increased. The results show that all participants significantly improved in their ability to use chopsticks in each intervention session. In addition, Assessment of Motor and Process Skills (AMPS) improved the generalization. According to the AMPS, both the overall motor and process skills increased from baseline an average of 0.7 logit. The results of this study showed occupational-based intervention on chopsticks skill to be effective in acquisition and generalization of chopstick skill in children with ASD.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.2
• /
• pp.148-154
• /
• 2021

Text-Independent speaker verification needs to extract text-independent speaker embedding to improve generalization performance. However, deep neural networks that depend on training data have the potential to overfit text information instead of learning the speaker information when repeatedly learning from the identical time series. In this paper, to prevent the overfitting, we propose a segment unit shuffling layer that divides and rearranges the input layer or a hidden layer along the time axis, thus mixes the time series information. Since the segment unit shuffling layer can be applied not only to the input layer but also to the hidden layers, it can be used as generalization technique in the hidden layer, which is known to be effective compared to the generalization technique in the input layer, and can be applied simultaneously with data augmentation. In addition, the degree of distortion can be adjusted by adjusting the unit size of the segment. We observe that the performance of text-independent speaker verification is improved compared to the baseline when the proposed segment unit shuffling layer is applied.

• Journal of Institute of Control, Robotics and Systems
• /
• v.6 no.3
• /
• pp.273-283
• /
• 2000

In this paper we suggest an optimal design method of Fuzzy-Neural Networks(FNN) model for complex and nonlinear systems. The FNNs use the simplified inference as fuzzy inference method and Error Back Propagation Algorithm as learning rule. And we use a HCM(Hard C-Means) Clustering Algorithm to find initial parameters of the membership function. The parameters such as parameters of membership functions learning rates and momentum weighted value is proposed to achieve a sound balance between approximation and generalization abilities of the model. According to selection and adjustment of a weighting factor of an aggregate objective function which depends on the number of data and a certain degree of nonlinearity (distribution of I/O data we show that it is available and effective to design and optimal FNN model structure with a mutual balance and dependency between approximation and generalization abilities. This methodology sheds light on the role and impact of different parameters of the model on its performance (especially the mapping and predicting capabilities of the rule based computing). To evaluate the performance of the proposed model we use the time series data for gas furnace the data of sewage treatment process and traffic route choice process.

• Communications for Statistical Applications and Methods
• /
• v.30 no.5
• /
• 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.