• Physical Therapy Korea
• /
• v.8 no.1
• /
• pp.89-96
• /
• 2001

This study addresses the effects of treadmill training on hyperextended knee and cadence in patients with hemiplegia. A single subject research design with multiple baselines across individuals was used for the study. Two patients with hemiplegia participated in the experiment. The experiment consisted of interventions where the patients were asked to ambulate for 15 minutes at a comfortable walking speed on the treadmill with 11% slope grade and were allowed to rest for 10 minutes. Patients, then, were asked to ambulated 20 meters at walkway. The number of occurrences of knee hyperextension and the total number of steps were recorded. The results showed that the occurrence of knee hyperextension decreased by approximately 30% after the first session of the treadmill training and continued to gradually decrease during the following sets of treadmill training. Meanwhile, there was a slight increase in the cadence to a negligible extent. These results suggest that the gait training on the sloped treadmill may be helpful for correcting the knee hyperextension in patients with hemiplegia.

• Journal of the Korean Society for information Management
• /
• v.23 no.4 s.62
• /
• pp.277-294
• /
• 2006

This paper examines a level of categorization performance in a real-life collection of abstract articles in the fields of science and technology, and tests the optimal size of documents per category in a training set using a kNN classifier. The corpus is built by choosing categories that hold more than 2,556 documents first, and then 2,556 documents per category are randomly selected. It is further divided into eight subsets of different size of training documents : each set is randomly selected to build training documents ranging from 20 documents (Tr-20) to 2,000 documents (Tr-2000) per category. The categorization performances of the 8 subsets are compared. The average performance of the eight subsets is 30% in $F_1$ measure which is relatively poor compared to the findings of previous studies. The experimental results suggest that among the eight subsets the Tr-100 appears to be the most optimal size for training a km classifier In addition, the correctness of subject categories assigned to the training sets is probed by manually reclassifying the training sets in order to support the above conclusion by establishing a relation between and the correctness and categorization performance.

• Journal of Multimedia Information System
• /
• v.6 no.4
• /
• pp.225-234
• /
• 2019

Deep learning has become increasingly popular in both academic and industrial areas nowadays. Various domains including pattern recognition, Computer vision have witnessed the great power of deep neural networks. However, current studies on deep learning mainly focus on quality data sets with balanced class labels, while training on bad and imbalanced data set have been providing great challenges for classification tasks. We propose in this paper a method of data analysis-based data reduction techniques for selecting good and diversity data samples from a large dataset for a deep learning model. Furthermore, data sampling techniques could be applied to decrease the large size of raw data by retrieving its useful knowledge as representatives. Therefore, instead of dealing with large size of raw data, we can use some data reduction techniques to sample data without losing important information. We group PCB characters in classes and train deep learning on the ResNet56 v2 and SENet model in order to improve the classification performance of optical character recognition (OCR) character classifier.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.27 no.8
• /
• pp.1239-1249
• /
• 1990

In this paper, using the neural network design techniques, an adaptive Mahalanobis distance classifier(AMDC) is designed. This classifier has three layers: input layer, internal layer and output layer. The connection from input layer to internal layer is fully connected, and that from internal to output layer has partial connection that might be thought as an Oring. If two ormore clusters of patterns of one class are laid apart in the feature space, the network adaptively generate the internal nodes, whhch are corresponding to the subclusters of that class. The number of the output nodes in just same as the number of the classes to classify, on the other hand, the number of the internal nodes is defined by the number of the subclusters, and can be optimized by itself. Using the method of making the subclasses, the different patterns that are of the same class can easily be distinguished from other classes. If additional training is needed after the completion of the traning, the AMDC does not have to repeat the trainging that has already done. To test the performance of the AMDC, the experiments of classifying 500 Hangeuls were done. In experiment, 20 print font sets of Hangeul characters(10,000 cahracters) were used for training, and with 3 sets(1,500 characters), the AMDC was tested for various initial variance \ulcornerand threshold \ulcorner and compared with other statistical or neural classifiers.

• Bioinformatics and Biosystems
• /
• v.2 no.1
• /
• pp.28-32
• /
• 2007

The protein phosphorylation is one of the important processes in the cell signaling pathway. A variety of protein kinase families are involved in this process, and each kinase family phosphorylates different kinds of substrate proteins. Many methods to predict the kinase-specific phosphoryrated sites or different types of phosphorylated residues (Serine/Threonine or Tyrosin) have been developed. We employed Supprot Vector Machine (SVM) to attempt the prediction of protein kinase specific phosphorylation sites. 10 different kinds of protein kinase families (PKA, PKC, CK2, CDK, CaM-KII, PKB, MAPK, EGFR) were considered in this study. We defined 9 residues around a phosphorylated residue as a deterministic instance from which protein kinases determine whether they act on. The subsets of PSI-BALST profile was converted to the numerical vectors to represent positive or negative instances. When SVM training, We took advantage of multiple SVMs because of the unbalanced training sets. Representative negative instances were drawn multiple times, and generated new traing sets with the same positive instances in the original traing set. When testing, the final decisions were made by the votes of those multiple SVMs. Generally, RBF kernel was used for the SVMs, and several parameters such as gamma and cost factor were tested. Our approach achieved more than 90% specificity throughout the protein kinase families, while the sensitivities recorded 60% on average.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.18 no.11
• /
• pp.47-56
• /
• 2019

A deep neural network (DNN) model was proposed to predict the upset in the inertia friction welding process using a database comprising results from a series of FEM analyses. For the database, the upset length, upset beginning time, and upset completion time were extracted from the results of the FEM analyses obtained with various of axial pressure and initial rotational speed. A total of 35 training sets were constructed to train the proposed DNN with 4 hidden layers and 512 neurons in each layer, which can relate the input parameters to the welding results. The mean of the summation of squared error between the predicted results and the true results can be constrained to within 1.0e-4 after the training. Further, the network model was tested with another 10 sets of welding input parameters and results for comparison with FEM. The test showed that the relative error of DNN was within 2.8% for the prediction of upset. The results of DNN application revealed that the model could effectively provide welding results with respect to the exactness and cost for each combination of the welding input parameters.

• Nuclear Engineering and Technology
• /
• v.41 no.9
• /
• pp.1181-1190
• /
• 2009

Knowing more about the Local Power Density (LPD) at the hottest part of a nuclear reactor core can provide more important information than knowledge of the LPD at any other position. The LPD at the hottest part needs to be estimated accurately in order to prevent the fuel rod from melting in a nuclear reactor. Support Vector Machines (SVMs) have successfully been applied in classification and regression problems. Therefore, in this paper, the power peaking factor, which is defined as the highest LPD to the average power density in a reactor core, was estimated by SVMs which use numerous measured signals of the reactor coolant system. The SVM models were developed by using a training data set and validated by an independent test data set. The SVM models' uncertainty was analyzed by using 100 sampled training data sets and verification data sets. The prediction intervals were very small, which means that the predicted values were very accurate. The predicted values were then applied to the first fuel cycle of the Yonggwang Nuclear Power Plant Unit 3. The root mean squared error was approximately 0.15%, which is accurate enough for use in LPD monitoring and for core protection that uses LPD estimation.

• Structural Engineering and Mechanics
• /
• v.70 no.6
• /
• pp.671-681
• /
• 2019

This paper examines the applicability of artificial neural network (ANN) and multivariate adaptive regression splines (MARS) to predict the compressive strength of bacteria incorporated geopolymer concrete (GPC). The mix is composed of new bacterial strain, manufactured sand, ground granulated blast furnace slag, silica fume, metakaolin and fly ash. The concentration of sodium hydroxide (NaOH) is maintained at 8 Molar, sodium silicate ($Na_2SiO_3$) to NaOH weight ratio is 2.33 and the alkaline liquid to binder ratio of 0.35 and ambient curing temperature ($28^{\circ}C$) is maintained for all the mixtures. In ANN, back-propagation training technique was employed for updating the weights of each layer based on the error in the network output. Levenberg-Marquardt algorithm was used for feed-forward back-propagation. MARS model was developed by establishing a relationship between a set of predictors and dependent variables. MARS is based on a divide and conquers strategy partitioning the training data sets into separate regions; each gets its own regression line. Six models based on ANN and MARS were developed to predict the compressive strength of bacteria incorporated GPC for 1, 3, 7, 28, 56 and 90 days. About 70% of the total 84 data sets obtained from experiments were used for development of the models and remaining 30% data was utilized for testing. From the study, it is observed that the predicted values from the models are found to be in good agreement with the corresponding experimental values and the developed models are robust and reliable.

• Korean Journal of Agricultural Science
• /
• v.50 no.3
• /
• pp.527-538
• /
• 2023

In this study, we aimed to apply multispectral imaging (713 - 920 nm, 10 bands) for maturity classification and recognition of oriental melons grown in hydroponic greenhouses. A total of 20 oriental melons were selected, and time series multispectral imaging of oriental melons was 7 - 9 times for each sample from April 21, 2023, to May 12, 2023. We used several approaches, such as Savitzky-Golay (SG), standard normal variate (SNV), and Combination of SG and SNV (SG + SNV), for pre-processing the multispectral data. As a result, 713 - 759 nm bands were preprocessed with SG for the maturity classification of oriental melons. Additionally, a Light Gradient Boosting Machine (LightGBM) was used to train the recognition model for oriental melon. R² of recognition model were 0.92, 0.91 for the training and validation sets, respectively, and the F-scores were 96.6 and 79.4% for the training and testing sets, respectively. Therefore, multispectral imaging in the range of 713 - 920 nm can be used to classify oriental melons maturity and recognize their fruits.

• Journal of Korea Water Resources Association
• /
• v.33 no.2
• /
• pp.247-262
• /
• 2000

In this paper, the WSANN(Water Stage Analysis with Neural Network) model was presented so as to predict flood water stage at Jindong which has been the major stream gauging station in Nakdong river basin. The WSANN model used the improved backpropagation training algorithm which was complemented by the momentum method, improvement of initial condition and adaptive-learning rate and the data which were used for this study were classified into training and testing data sets. An empirical equation was derived to determine optimal hidden layer node between the hidden layer node and threshold iteration number. And, the calibration of the WSANN model was performed by the four training data sets. As a result of calibration, the WSANN22 and WSANN32 model were selected for the optimal models which would be used for model verification. The model verification was carried out so as to evaluate model fitness with the two-untrained testing data sets. And, flood water stages were reasonably predicted through the results of statistical analysis. As results of this study, further research activities are needed for the construction of a real-time warning of the impending flood and for the control of flood water stage with neural network method in river basin. basin.