|
Classification and Visualization of Long-Term Life-monitoring Sensor Signals Using Topological Characteristics of Category Maps |
|
Kazuya Iguchi, Hirokazu Madokoro, Kazuhito Sato(Akita Prefectural University, Japan), Kazuhisa Nakasho(Yamaguchi University, Japan), Nobuhiro Shimoi(Akita Prefectural University, Japan) |
| This paper presents a novel extraction and visualization method of human behavior patterns as life rhythms from sensor signals obtained using our originally developed life-monitoring system. Our method visualizes categorical relations and distribution characteristics on category maps and their fired units. Experimentally obtained results reveal that the distribution of burst units is spread evenly along with the setting of learning iterations greater than the data size. This characteristic indicates that it is necessary to increase learning iterations when the mapping size is increased.
|
|
|
Document Classification Based on KNN Algorithm by Term Vector Space Reduction |
|
Aiman Nickolayevna Moldagulova(International Information Technology University, Kazakhstan), Rosnafisah Bte. Sulaiman(Universiti Tenaga Nasional, Malaysia) |
| Nowadays there is an increasing interest in the area of unstructured data analysis. The vast majority of unstructured data belongs to unstructured text data. Retrieving useful information from huge volume of unstructured text data is very challenging task. Text mining is a thought-provoking research area as it tries to discover knowledge from unstructured text. This paper deals with methods used for handling unstructured text data in particular document classification problems. Most document classification methods based on term vector space model of representation of unstructured textual data. |
|
|
Suitable Activity Function of Neural Networks for Data Enlargement |
|
Isaac Job Betere, Hiroshi Kinjo, Kunihiko Nakazono, Naoki Oshiro(University of the Ryukyus, Japan) |
| In this paper, we present a study on activity functions for a multi-layered neural networks (MLNNs) and propose a suitable activity function for data enlargement (DE). We have carefully studied the training performance of Sigmoid, ReLu, Leaky-ReLu and L & exp. activity functions for three inputs to multiple output training patterns. Our MLNNs model has L hidden layers with two inputs to four or six outputs by backpropagation neural network training (BP). We focused on the multi teacher training signals to investigate and evaluate the training performance in MLNNs and select the best and good a |
|
|
Residue detection in the large intestine from colonoscopy video using the support vector machine method |
|
Minwoo Cho(Seoul National University, Korea), Hyoun Joong Kong(Chungnam National University, Korea), Jee Hyun Kim(Seoul National University Boramae Medical Center, Korea), Byoungjun Jeon(Seoul National University, Korea), Kyoung Sup Hong(Mediplex Sejong Hospital, Korea), Sungwan Kim(Seoul National University, Korea) |
| When performing colonoscopy or applying image processing to colonoscopy video, the presence of residue in the large intestine is negative affecting factor. In this reason, the colonoscopist evaluates and records the colon cleanliness after performing colonoscopy. However, these assessments can be influenced by the
subjectivity of the colonoscopist. To quantify the frequency of residues in the large intestine, we applied image processing and machine learning techniques to the digital images obtained from colonoscopy. |
|
|
Finding High Accuracy Neural Network for Welding Defects Classification Using Efficient Neural Architecture Search via Parameter Sharing |
|
Min-Guk Kang, Dong-Joong Kang(Pusan National University, Korea) |
| In this paper, we test and evaluate a method to select the novel convolutional neural network to determine whether the architecture search method is effective for the welding defect images. The method is based on using Efficient Neural Architecture Search via parameter sharing(ENAS)[2]. Using ENAS, we were able to find an architecture that achieved 0% error for 1,322 test images. Also, in the case of the MNIST dataset, we could find a novel architecture that achieved 99.77% accuracy for 10,000 test images. |
| |