• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.4 s.304
• /
• pp.95-102
• /
• 2005

In this paper, we propose a user friendly object-based image retrieval system using the interaction between cortex and hippocampus. Most existing ways of queries in content-based image retrieval rely on query by example or query by sketch. But these methods of queries are not adequate to needs of people's various queries because they are not easy for people to use and restrict. We propose a method of automatic color object extraction using CSB tree map(Color and Spatial based Binary をn map). Extracted objects were transformed to bit stream representing information such as color, size and location by region labelling algorithm and they are learned by the hippocampal neural network using the interplay between cortex and hippocampus. The cells of exciting at peculiar features in brain generate the special sign when people recognize some patterns. The existing neural networks treat each attribute of features evenly. Proposed hippocampal neural network makes an adaptive fast content-based image retrieval system using excitatory learning method that forwards important features to long-term memories and inhibitory teaming method that forwards unimportant features to short-term memories controlled by impression.

• Asian-Australasian Journal of Animal Sciences
• /
• v.31 no.11
• /
• pp.1700-1713
• /
• 2018

Objective: The aim of this study was to verify the usefulness of artificial neural networks (ANN), multivariate adaptive regression splines (MARS), naïve Bayes classifier (NBC), general discriminant analysis (GDA), and logistic regression (LR) for dystocia detection in Polish Holstein-Friesian Black-and-White heifers and cows and to indicate the most influential predictors of calving difficulty. Methods: A total of 1,342 and 1,699 calving records including six categorical and four continuous predictors were used. Calving category (difficult vs easy or difficult, moderate and easy) was the dependent variable. Results: The maximum sensitivity, specificity and accuracy achieved for heifers on the independent test set were 0.855 (for ANN), 0.969 (for NBC), and 0.813 (for GDA), respectively, whereas the values for cows were 0.600 (for ANN), 1.000 and 0.965 (for NBC, GDA, and LR), respectively. With the three categories of calving difficulty, the maximum overall accuracy for heifers and cows was 0.589 (for MARS) and 0.649 (for ANN), respectively. The most influential predictors for heifers were an average calving difficulty score for the dam's sire, calving age and the mean yield of the farm, where the heifer was kept, whereas for cows, these additionally included: calf sex, the difficulty of the preceding calving, and the mean daily milk yield for the preceding lactation. Conclusion: The potential application of the investigated models in dairy cattle farming requires, however, their further improvement in order to reduce the rate of dystocia misdiagnosis and to increase detection reliability.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.

• Journal of Intelligence and Information Systems
• /
• v.20 no.1
• /
• pp.35-48
• /
• 2014

According to the 2013 construction market outlook report, the liquidation of construction companies is expected to continue due to the ongoing residential construction recession. Bankruptcies of construction companies have a greater social impact compared to other industries. However, due to the different nature of the capital structure and debt-to-equity ratio, it is more difficult to forecast construction companies' bankruptcies than that of companies in other industries. The construction industry operates on greater leverage, with high debt-to-equity ratios, and project cash flow focused on the second half. The economic cycle greatly influences construction companies. Therefore, downturns tend to rapidly increase the bankruptcy rates of construction companies. High leverage, coupled with increased bankruptcy rates, could lead to greater burdens on banks providing loans to construction companies. Nevertheless, the bankruptcy prediction model concentrated mainly on financial institutions, with rare construction-specific studies. The bankruptcy prediction model based on corporate finance data has been studied for some time in various ways. However, the model is intended for all companies in general, and it may not be appropriate for forecasting bankruptcies of construction companies, who typically have high liquidity risks. The construction industry is capital-intensive, operates on long timelines with large-scale investment projects, and has comparatively longer payback periods than in other industries. With its unique capital structure, it can be difficult to apply a model used to judge the financial risk of companies in general to those in the construction industry. Diverse studies of bankruptcy forecasting models based on a company's financial statements have been conducted for many years. The subjects of the model, however, were general firms, and the models may not be proper for accurately forecasting companies with disproportionately large liquidity risks, such as construction companies. The construction industry is capital-intensive, requiring significant investments in long-term projects, therefore to realize returns from the investment. The unique capital structure means that the same criteria used for other industries cannot be applied to effectively evaluate financial risk for construction firms. Altman Z-score was first published in 1968, and is commonly used as a bankruptcy forecasting model. It forecasts the likelihood of a company going bankrupt by using a simple formula, classifying the results into three categories, and evaluating the corporate status as dangerous, moderate, or safe. When a company falls into the "dangerous" category, it has a high likelihood of bankruptcy within two years, while those in the "safe" category have a low likelihood of bankruptcy. For companies in the "moderate" category, it is difficult to forecast the risk. Many of the construction firm cases in this study fell in the "moderate" category, which made it difficult to forecast their risk. Along with the development of machine learning using computers, recent studies of corporate bankruptcy forecasting have used this technology. Pattern recognition, a representative application area in machine learning, is applied to forecasting corporate bankruptcy, with patterns analyzed based on a company's financial information, and then judged as to whether the pattern belongs to the bankruptcy risk group or the safe group. The representative machine learning models previously used in bankruptcy forecasting are Artificial Neural Networks, Adaptive Boosting (AdaBoost) and, the Support Vector Machine (SVM). There are also many hybrid studies combining these models. Existing studies using the traditional Z-Score technique or bankruptcy prediction using machine learning focus on companies in non-specific industries. Therefore, the industry-specific characteristics of companies are not considered. In this paper, we confirm that adaptive boosting (AdaBoost) is the most appropriate forecasting model for construction companies by based on company size. We classified construction companies into three groups - large, medium, and small based on the company's capital. We analyzed the predictive ability of AdaBoost for each group of companies. The experimental results showed that AdaBoost has more predictive ability than the other models, especially for the group of large companies with capital of more than 50 billion won.