Recently a pencil model has attracted attention again in view of behavioral approach. It is well known that a pencil can be transformed into a canonical form, but the traditional numerical computation method is implemented in terms of pivoting operation. In this paper, we make it clear what kind of mode the invariant subspace corresponds to, and propose a new algorithm to transform the pencil into the quasi-diagonal form. Since the method is based on geometric approach, it has an advantage of numerical stability.
In this paper, we show that a value function under risk is useful to model low probability and high consequence damage events like an earthquake for which expected utility theory is inadequate. Firstly, we assume alternatives to improve buildings, some scenarios of earthquakes, costs to improve buildings, probability of death and injury and cost of restoring building's damage for each scenario. Then, we show that the value function under risk is an appropriate approach to model and analyze decision making process with low probability and high consequence events.
In this paper we discuss training of three-layered neural network classifiers by solving inequalities. Namely, first we represent each class by the center of the training data belonging to the class, and determine the set of hyperplanes that separate each class (i.e., each center) into a single region. Then according to whether the center is on the positive or negative side of the hyperplane, we determine the target values of each class for the hidden neurons (i.e., hyperplanes). Since the convergence condition of the neural network classifier is now represented by the two sets of inequalities, we solve the sets successively by the Ho-Kashyap algorithm. We demonstrate the advantage of our method over the backpropagation algorithm using several benchmark data sets.
This paper proposes a new method for measuring the position and attitude of an autonomous land vehicle navigating in a rough terrain. The measurement system proposed here provides precise information on position and heading quickly and continuously. It consists of two laser scanners mounted on the vehicle and corner cubes placed in the environment as landmarks. Each laser scanner rotates a fan-shaped laser beam for detecting the retro-reflections by the corner cubes and measures their azimuth angles. This paper presents the principle of measurement and method of positional and attitude estimation. Simulation results are given to show the accuracy of the proposed method. This paper reports the measurement principle, simulation and experimental results.
We characterize a probabilistic measure named Model Set Unfalsified Probability (MSUP) for model set validation, where the model set is described by an LFT (Linear Fractional Transformation) form. We derive upper and lower bounds of MSUP and show that the lower bound computation can be reduced to an LMI-based convex optimization. A necessary and sufficient condition for which MSUP=0.5 (50%) is also provided. A numerical example confirms that the probabilistic approach more appropriately evaluates the suitability of a model set in robust controller design than deterministic approaches.