Keiichi Horio

FEEDBACK SELF-ORGANIZING MAP AND ITS APPLICATION TO SPATIO-TEMPORAL PATTERN CLASSIFICATION

In this paper, a feedback self-organizing map (FSOM), which is an extension of the self-organizing map (SOM) by employing feedback loops, is proposed. The SOM consists of an input layer and a competitive layer, and the input vectors applied to the input layer are mapped to the competitive layer keeping their spatial features. In order to embed the temporal information to the SOM, feedback loops from the competitive layer to the input layer are employed. The winner unit in the competitive layer is not assigned by only current input vector but also past winner units, thus the temporal information can be embedded. The effectiveness and validity of the proposed FSOM are verified by applying it to a spatio-temporal pattern classification.

Self-organizing map hardware accelerator system and its application to realtime image enlargement

We propose a new fast learning algorithm for SOM and its digital hardware design based on the massively parallel architecture. When this proposed algorithm is realized by using Xilinx XC2V6000-6 FPGA, a maximum performance of 17500 MCUPS is achieved and up to 256 competing units (16 /spl times/ 16 map) can be implemented. Each competing unit have a weight vector which is represented by 128 elements of 16 bits accuracy. Furthermore, we applied the proposed hardware to a realtime digital image enlargement system. In the case of full color (24 bits) image enlargement from QQVGA (160 /spl times/ 120 pixel) to QVGA (320 /spl times/ 240 pixel), a proposed hardware requires only 0.12 second per image, while the personal computer (Intel XEON, 2.8 GHz Dual) requires more than 5 seconds per image.

Pattern Recognition Based on Relative Position of Local Features Using Self-Organizing Map

In this paper, we propose a new pattern recognition method based on relative position of local feature. In the visual system of human, the local features such as lines and curves are extracted, and they are integrated. In the proposed method, relative position of the gazing points which include local features are extracted using self-organizing map. A template matching concerning the local features is used for recognizing the patterns. The effectiveness of the proposed method is verified by applying it to MNIST handwritten digits database

The self-organizing relationship (SOR) network employing fuzzy inference based heuristic evaluation

When human beings acquire a new skill, this usually is accomplished by the summarization of numerous experiences based on their own evaluation criteria. Usually these experiences are obtained by trial and error. The criteria for success and failure are based on our own knowledge or advice given by others. The Self-Organizing Relationship (SOR) network has been inspired by this process and has been proposed to emulate this process computationally. In the previous applications of the SOR network for controller design, the evaluation criteria have been assigned by using mathematical expressions. Generally, however, mathematical expressions of the evaluation criteria become difficult as the complexity of a target system increases. On the other hand, human beings can contrive to express their knowledge for evaluation by using heuristic expressions, although a target system is complicated. In this study, we employ fuzzy inference in order to realize heuristic expressions of the evaluation criteria.

Effective hierarchical optimization by a hierarchical multi-space competitive genetic algorithm for the flexible job-shop scheduling problem

We propose an effective hierarchical optimize method HmcDGA.The HmcDGA not only optimizes individuals, but it also optimizes solution space.The HmcDGA can find an optimal solution at low computational cost.The HmcDGA does not require special operations for specific problems.We apply the proposed HmcDGA to the FJSP, and evaluate its effectiveness. In this paper, we propose a new optimization technique, the hierarchical multi-space competitive distributed genetic algorithm (HmcDGA), which is effective for the hierarchical optimization problem. It is an extension of the multi-space competitive distributed genetic algorithm (mcDGA), which was proposed by the authors. The mcDGA efficiently finds an optimal solution with a low computational cost by increasing the number of individuals in a solution space in which it is likely to exist. An optimization method that is divided into several levels of hierarchy is called a hierarchical optimization. Several hierarchical optimization techniques have been proposed, including the hierarchical genetic algorithm (HGA). In hierarchical optimization, a complex problem is divided into a hierarchical collection of simpler problems, and each level is optimized independently. In this way, complex problems can be solved without the need to develop problem-specific operators. However, in the conventional HGA, this results in a high computational cost because the genetic algorithm (GA) is repeated many times at upper and lower level. The HmcDGA is a hybrid of the mcDGA and HGA, and it has some of the advantages of each one; for example, the HmcDGA can find an optimal solution at low computational cost and without requiring special operations. This allows it to be applied to a wide variety of optimization problems. Therefore, the HmcDGA may become the powerful optimization algorithm that can solve various problems. In this paper, we apply the proposed HmcDGA to the flexible job-shop scheduling problem (FJSP) which is one of the complex combinational optimization problem and confirm its effectiveness. Simulation results show that the HmcDGA can find solutions that are comparable to those found by using GAs developed specifically for the FJSP, the HmcDGA is not required a lot of computational costs comparing to the HGA.

Advanced self-organizing maps using binary weight vector and its digital hardware design

Many co-processors which are designed for learning of self-organizing maps (SOM) have been proposed in order to reduce the processing time. However, hardware in which all processes of the learning of the SOM are achieved is not realized, because it needs many complex calculations. In this study, a new learning algorithm of the SOM in which input vectors and weight vectors are represented as binary data is proposed. The effectiveness of the proposed algorithm is verified by designing the digital hardware of the proposed algorithm using HDL.

Self-organizing map with input data represented as graph

This paper proposes a new method of Self-Organizing Map (SOM) in which an input space is represented as a graph by modifications of a distance measure and a updating rule. The distance between input node and reference element is defined by the shortest distance between them in the graph. The reference elements are updated along the shortest path to the input node. The effectiveness of the proposed method is verified by applying it to a Traveling Salesman Problem.

A bit-shifting-based fuzzy inference for self-organizing relationship (SOR) network

We propose a bit-shifting-based fuzzy inference method for an efficient digital hardware implementation. The proposed fuzzy inference method includes two new techniques which are a membership function generating method and a fast defuzzification method using only “active units”. These techniques reduce a hardware cost and a calculation cost for the membership function and the defuzzification, respectively. In this paper, we apply the proposed method to an execution mode of self-organizing relationship network. Simulation results show that the proposed method has a good approximation ability of a nonlinear I/O relationship as well as the ordinary method.

Adaptive self-organizing relationship network and its application to adaptive control

An adaptive self-organizing relationship (ASOR) network, which is the extension of the self-organizing relationship (SOR) network proposed by the authors, is proposed. The SOR network can obtain the desired input/output relationship of a target system by using the input/output vector pairs and their evaluations. In order to add the ability of adaptation to the SOR network, the new algorithm that the learning rate and the area of the neighborhood are adjusted according to need is employed. The ASOR network can adapt to the change of the desired input/output relationship of the target system. The effectiveness of the proposed ASOR network is verified by applying it to design of the control system of the DC motor whose load changes with time.

The Cognitive Mechanisms of Multi-scale Perception for the Recognition of Extremely Similar Faces

We aimed to examine the cognitive question of why human observers have difficultly in distinguishing two extremely similar faces of people of different genders, by using hybrid images (HIs). For this purpose, we proposed a computational model of the cognitive processing in the brain of multi-scale perception, which incorporates two different roles of high- and low-spatial scales in face recognition. This model was based on the multi-scale correspondence between the sizes of the filters and images in the Gabor pyramid. Multi-scale correspondence with relatively small Gabor kernels demonstrated that the scale similarity curves were qualitatively consistent with the gain functions for spatial frequency in the Gaussian filters. Locally high-scale similarities for both the low- and the high-pass filtered images indicated that the face in the HI was misrecognized as the face in the counterpart filtered image. Smaller-scale similarity differences with greater spatial frequency overlap of the low- and high-pass filters of the HI demonstrated that human observers fail to employ appropriate combinations of high- and low-scale representations to distinguish extremely similar faces. The larger Gabor kernels suggested that human observers fail to identify the face in the HI at the low resolution.