A freehand curve identification method termed fuzzy spline curve identifier (FSCI) was proposed. FSCI identifies a drawn stroke as one of seven geometric curves (line, circle, circular arc, ellipse, elliptic arc, closed free curve, and opened free curve) based on the shape of the drawn curve and the fuzziness (or coarseness) of the drawing manner. Then, a snapping method termed multi-resolution fuzzy grid snapping (MFGS) was proposed. MFGS shapes and aligns the geometric curves identified by FSCI according to a given square grid system. Then, a sketch-based interface for CAD systems was realized by the combination of FSCI and MFGS, where a user can arrange the seven kinds of geometric curves and complete a variety of geometric drawings just by sketching. When a circular arc or an elliptic arc is identified by FSCI in the interface, its parameters such as central angle, major axis length, and minor axis length are quantized to fit the grid system before it is snapped by MFGS. However, in cases where the identified arcs have shallow shapes, the quantization rarely works suitably. Hence, MFGS frequently fails in shaping and aligning such arcs to fit the grid system. In this study, we first show that the shaping and aligning of circular arcs and elliptic arcs with parameter quantization becomes essentially difficult in numerous cases. Then, to avoid this problem, we propose a new technique termed sub-curve identification by modifying the curve identification of FSCI. The proposed technique identifies the “n-quarter circular arcs” and “n-quarter elliptic arcs” that are appropriately snapped by MFGS without parameter quantization. Finally, we realize a new sketch-based interface where a user can arrange n-quarter circular arcs, n-quarter elliptic arcs, and the seven kinds of geometric curves and experimentally demonstrate that it effectively works as a CAD interface.
Human have two types of inferences: intuitive and logical. In traditional inference studies, the intuitive inference has modeled by a probabilistic method like the Bayesian, and the logical inference has modeled by a symbolic method like the Tree search. There are many studies that relate inference behavior with brain areas, but few have focused on the mechanism of logical inference that can emerge naturally from the brain neural circuit. So, in this study, we assumed that the human inference process should not be divided into intuitive and logical ones, but should be modeled as an operating mode switching of a single distributed neural network. We used an associative memory model for its verification. The intuitive inference function is realized by a combination of a memory association from a current state and an association of the memory state with the value recognition. Then, the logical inference like behavior is realized by repeatedly biasing the gain of the valued memory state found in the intuitive inference process. A computer simulation in a maze search task is conducted, and we confirmed the emergence of the symbolic tree search like inference behavior with pruning of low probability branch from the intuitive like probabilistic inference by the change of the calculation parameter of the same model.
In this study we estimated the understanding of words in order to visualize a person’s individual vocabulary. We have hypothesized that words which people understand or, conversely, do not, tend to be concentrated on certain word concepts (concepts indicated by the word). Therefore, we proposed a method for estimating the understanding of words combining word concept created by word2vec and Hierarchical Neural Network, and succeeded in estimating the understanding of words with an accuracy of over 72%. From the results, it was suggested that the validity of the hypothesis and the word concept were useful for estimating the understanding of words. With implementation of additional indicators such as difficulty, estimation at a practical level can be anticipated.
Long-term time series data can be understood according to transitions of trends and features. We must partition time series into several terms of intervals of different trends and features. In statistics, change detection expresses it as a degree of change using probability distributions. In this paper, we propose a method to partition time series using a hierarchical clustering. First, we have clusters of one line segment connecting adjacent data. Then, we merge two adjacent similar clusters into one with a total similarity calculated by the weighted average of three similarities of values, change of values and oscillations. However, the partition results with the fixed weights do not fit our sense. We, therefore, propose variable weights with three similarities and sizes of adjacent clusters. Furthermore, in order to exclude small clusters including outliers, we define a similarity of two clusters adjacent to the small cluster. We apply this method to actual time series and show results.
In this paper, we propose an agent which selects utterance strategy according to driver’s attributes and situation based on politeness theory. The proposed agent system considers attribute information such as age, gender, personality, driving experience and driving characteristics of the driver, and selects and supports highly receptive utterance strategies. Here, as a survey for the development of the proposed system, an impression evaluation experiment on a driving support agent was conducted using a moving picture reflecting driving support scenes. In particular, we focused attention on the difference of the end-of-sentence style which is a representation method of distinctive psychological distances in Japanese conversation. As a result of the experiment, it is suggested that agents using non-polite words may be effective for improving familiarity. On the other hand, the agent using honorifics gives the impression that it is cautious, suggesting the possibility that it feels that information is conveyed accurately.