Noritaka Shigei

Performance comparison of hybrid electromagnetism-like mechanism algorithms with descent method

Abstract Electromagnetism-like Mechanism (EM) method is known as one of metaheuristics. The basic idea is one that a set of parameters is regarded as charged particles and the strength of particles is corresponding to the value of the objective function for the optimization problem. Starting from any set of initial assignment of parameters, the parameters converge to a value including the optimal or semi-optimal parameter based on EM method. One of its drawbacks is that it takes too much time to the convergence of the parameters like other meta-heuristics. In this paper, we introduce hybrid methods combining EM and the descent method such as BP, k-means and FIS and show the performance comparison among some hybrid methods. As a result, it is shown that the hybrid EM method is superior in learning speed and accuracy to the conventional methods.

Bagging and AdaBoost algorithms for vector quantization

In this paper, we propose VQ methods based on ensemble learning algorithms Bagging and AdaBoost. The proposed methods consist of more than one weak learner, which are trained in parallel or sequentially. In Bagging, the weak learners are trained in parallel by using randomly selected data from a given data set. The output for Bagging is given as the average among the weak learners. In AdaBoost, the weak learners are sequentially trained. The first weak learner is trained by using randomly selected data from a given data set. For the second and later weak learners, the probability distribution of learning data is modified so that each weak learner focuses on data involving higher error for the previous weak one. The output for AdaBoost is given as the weighted average among the weak learners. The presented simulation results show that the proposed methods can achieve a good performance in shorter learning times than conventional ones such as K-means and NG.

A Proposal of a Quantum Search Algorithm

For searching an unsorted database with N items, a classical computer requires O(N) steps but Grovers quantum searching algorithm requires only O(√N) steps. However, it is also known that Grovers algorithm is effective in the case where the initial amplitude distribution of dataset is uniform, but is not always effective in the non-uniform case. In this paper, we will consider the influence of initial amplitude distribution of dataset for Grovers and Venturas algorithms. Further, in order to improve the performance of quantum searching, we propose a new algorithm and analyze its dynamics.

Some quantum search algorithms for arbitrary initial amplitude distribution

For searching any item in an unsorted database with N items, Grovers quantum searching algorithm takes only O(√N) steps, which is much faster than O(N) steps required for a classical computer. However, it is known that Grovers algorithm is effective only in the case where the initial amplitude distribution of dataset is uniform. On the other hand, Ventura has also proposed a quantum searching algorithm. The algorithm is effective only in the special case for the initial amplitude distribution. Therefore, we have proposed an effective quantum searching algorithm in another case in the previous paper. In this paper, we generalize the previous algorithm and propose several effective algorithms in several cases where initial amplitude distributions of dataset are non-uniform. Further, in order to show the effectiveness of the algorithms, we analyze their dynamics.

Effective initial route construction for mobile relay on wireless sensor network

In recent years, mobile relay has been studied in order to reduce the energy consumption in WSN El-Moukaddem et al. (Mobile Relay Configuration in Data-intensive Wireless Sensor Networks, 2009). The concept of mobile relay is that some movable nodes change their location so as to minimize the total energy consumed by both wireless transmission and locomotion. Mobile relay needs to determine an initial route that represents the sequence of relaying nodes. We have already proposed several initial route construction algorithms based on greedy approaches Shigei et al. (IAENG Int J Comput Sci 39: 321–328, 2012), Yudo et al. (Battery-Aware Initial Route Construction for Mobile Relay on Wireless Sensor Network, 2012). However, the conventional method cannot always provide optimal routes, because they do not examine all the possible routes. In this paper, we propose a battery-aware initial route construction method based on Dijkstra’s algorithm (BAIR-D), which utilizes the node information outside of the direct communication range. The algorithm has a capability to examine all the possible routes. We show the effectiveness by numerical simulation.

A multiple vector quantization approach to image compression

This paper investigates the effectiveness of a parallelized approach to VQ based image compression. In particular, we consider an image compression method using multiple VQs. The method, called MVQ, generates multiple independent codebooks to compress an image by using a neural network algorithm. In the image restoration, MVQ restores low quality images from the multiple codebooks, and then combines the low quality ones into a high quality one. Further, we present an effective coding scheme for codebook indexes to overcome the in-efficiency of MVQ in compression rate. Our simulation results show that the MVQ method outperforms a conventional single-VQ method when the compression rate is smaller than some values.

Higher order differential correlation associative memory of sequential patterns

This paper describes some properties of storage capacity and robustness of differential correlation associative memory of sequential patterns using higher order neural networks. First, it is shown that storage capacities for k=1, 2 and 3 dimensional cases are 0.059N, 0.023(/sub 2//sup N/) and 0.014(/sub 3//sup N/) from the prediction using the transition properties, respectively, where N is the number of neurons and (/sub K//sup N/) means the combination of k from N. And it is shown that higher order models are superior in the pattern selection ability to the conventional one. Further, it is shown that higher order differential correlation models have high robustness compared to the conventional correlation models.

A Proposal of Fuzzy Inference Model Composed of Small-Number-of-Input Rule Modules

The automatic construction of fuzzy system with a large number of input variables involves many difficulties such as large time complexities and getting stuck in a shallow local minimum. In order to overcome them, an SIRMs (Single-Input Rule Modules) model has been proposed. However, such a simple model does not always achieve good performance in complex non-linear systems. This paper proposes a fuzzy reasoning model as a generalized SIRMs model, in which each module deals with a small number of input variables. The reasoning output of the model is determined as the weighted sum of all modules, where each weight is the importance degree of a module. Further, in order to construct a simpler model, we introduce a module deletion function according to the importance degree into the proposed system. With the deletion function, we propose a learning algorithm to construct a fuzzy reasoning system consisting of small-number-of-input rule modules (SNIRMs). The conducted numerical simulation shows that the proposed method is superior in terms of accuracy compared to the conventional SIRMs model.

Neural Networks Determining Injury by Salt Water for Distribution Lines

Japan has so may solitary islands and wide coastal areas. In the air of these areas, there are so many grains of sea-salt. The salt grains adhere to distribution lines and damage the lines. In particular, the damage of covered electric wires is a serious problem. However, any method has not been proposed that judges if distribution lines are injured by salt water. This paper proposes a neural network method to determine injury of distribution lines by salt.

A proposal of privacy preserving reinforcement learning for secure multiparty computation

Many studies have been done with the security of cloud computing. Though data encryption is a typical approach, high computingxa0complexity for encryption and decryption of data is needed. Therefore, safe system for distributed processing with secure dataxa0attracts attention, and a lot of studies have been done. Secure multiparty computation (SMC) is one of these methods. Specifically,xa0two learning methods for machine learning (ML) with SMC are known. One is to divide learning data into several subsets andxa0perform learning. The other is to divide each item of learning data and perform learning. So far, most of works for ML with SMCxa0are ones with supervised and unsupervised learning such as BP and K-means methods. It seems that there does not exist anyxa0studies for reinforcement learning (RL) with SMC. This paper proposes learning methods with SMC for Q-learning which is onexa0of typical methods for RL. The effectiveness of proposed methods is shown by numerical simulation for the maze problem.