There are many occasions to read electrical texts along with the growth of computers and the Internet. Therefore, it is required to have an environment for selecting enormous information or for comprehending texts quickly. Automatic summarization methods are generally used to find or select information quickly. However, indicative summaries used in results of search engines do not express whole texts, so we cannot understand a topic is treated as one of the subjects of a text or is just partially appeared. On the other hand, informative summaries also require much time to grasp the contents of a text, because the summarization rates are not so high in order to summarize the whole text. In this paper, a system that expresses text coherence along with the topic words, the subject of a text, is described. All words in a text are classified into topic-related or not, and all are displayed in a two dimensional interface of the proposed river rafting system. This system is also applied to extract sentences that express conclusions of a text by using words repeatedly appearing related to the topic words.

View full abstract

Semantic lexicons, such as Roget's Thesaurus or WordNet, act as useful knowledge resources in natural language processing applications. However, such manually created lexical resources do not always reflect the new terms and named entities frequently found in the Web. Moreover, manually maintaining lexical resources are costly and time consuming. Motivated by this challenge, we propose a method to automatically extract related terms using the web as a corpus. The proposed method exploits snippets retrieved from a web search engine and efficiently finds related terms.

View full abstract

In this paper we propose an approach of predicting individual's sub-health based on the principle of TCM as a preventive medicine. The object's vision features like features of tongue, eye and face are extracted for modeling a process of TCM doctor's diagnosis. Because of the diversity and uncertainty of TCM doctors' diagnosis, the sensitivity is defined as a criterion to select the training data from the derived features and the diagnosis data given by different TCM doctors for constructing the sub-health inference model. The experiment results show that the sensitivity-based data selection improves the model's inference performance on the accuracy, correlation and residual variance.

View full abstract

We have proposed “biomodeling system”, in which the “top-down bio-processing” for sending actuator signals to robot from living neuronal network cultured on a 2-dimensional electrode arrays, and the “bottom-up robot-processing” for electrical stimulation to living neuronal network from robot are connected between neuronal network and robot. In this paper, we discuss two kinds of learning mechanism, which are the learning process of living neural network and the adaptability learning of fuzzy logic using the tracking estimation of Khepera II in a straight lane.

View full abstract

In recent years, Complex Networks have attracted much attention in research field. As a typical model, there are Small-world network and Scale-free network. In this research, we propose a method to configure growing network based on similarities. The proposed method refers to Self-Organizing Map algorithm. It aims to realize the generation of networks which show characteristics of the real network in a natural way. Generated network by the proposed method shows small characteristics path length (L), large clustering coefficient (C) and the power-law degree distribution. In addition, the attack of the high degree nodes, the network generated by the proposed method had little effect, indicating that the network was robust as seen in real world.

View full abstract

Fuzzy c-Means based Classifier (FCMC) is a simple approach to classification based on the clustering and parameter optimization methods. In a classifier design, training of the classifier takes a long time when the size of the training set is very large. The training time of FCMC is improved by reducing memory usage and by revising the random search approach. This paper reports the results of the comparison between FCMC and the state of the art classifier: LibSVM. The number of clusters of FCMC is increased up to a maximum of 2⁸=256. The two parameters out of four are automatically optimized by the revised random search approach. When the number of training samples is more than a million, the total training time for FCMC is estimated to be two to three orders of magnitude smaller than LibSVM, though FCMC achieves the same level of classification accuracy with LibSVM.

View full abstract