Akihiko Uchibori

Artificial life system for optimization of nonconvex functions

This paper presents a distributed algorithm for optimization of nonconvex multimodal functions. In recent years, new distributed algorithms based on artificial life (ALife) system has been studied and its potential power has been demonstrated. In this paper, the frame work of ALife system is employed into a function minimization. We also propose a hybrid algorithm in which ALife system is incorporated with the local search method for finding good start points for the local search. Since the proposed method utilizes no gradient information it can be applied to very wide class of optimization problems. The effectiveness of the proposed method is demonstrated through some numerical tests.

A design of non-linear electro-pneumatic servo systems by using neural-net based linearizer with off-line learning algorithm

Many electro-pneumatic servo systems have strong nonlinear property because of compressible air in cylinders. To control their systems, a neural network(NN) based linearizer with off line learning is proposed. The augmented system which consists of the controlled object and the linearizer is controlled by a linear controller. The point of the learning algorithm of the NN is to estimate from the actual input of the controlled object to the input of the augmented object directly.