Takayasu Fuchida

Lossless fractal image coding

Lossless fractal image coding (LFIC) is a new method to code images. The significant property of this method is that it is the first simple fractal image coding method that can regenerate images without data loss (lossless). The method, different from the other fractal image compression, needs no search for the matched domain-range. Therefore, the coding time of this method is also very fast compared with the other fractal image compression methods.

A proposition of adaptive state space partition in reinforcement learning with Voronoi tessellation

This paper presents a new adaptive segmentation of continuous state space based on vector quantization algorithm such as Linde–Buzo–Gray for high-dimensional continuous state spaces. The objective of adaptive state space partitioning is to develop the efficiency of learning reward values with an accumulation of state transition vector in a single-agent environment. We constructed our single-agent model in continuous state and discrete actions spaces using Q-learning function. Moreover, the study of the resulting state space partition reveals a Voronoi tessellation. In addition, the experimental results show that this proposed method can partition the continuous state space appropriately into Voronoi regions according to not only the number of actions, but also achieve a good performance of reward-based learning tasks compared with other approaches such as square partition lattice on discrete state space.

Meet-in-the-middle Attack with Splice-and-Cut Technique on the 19-round Variant of Block Cipher HIGHT

We show a meet-in-the-middle (MITM) attack with Splice-and-Cut technique (SCT) on the 19-round variant of the block cipher HIGHT. The original HIGHT having 32-round iteration was proposed by Hong et al. in 2006, which applies the 8-branch Type-2 generalized Feistel network (GFN) with 64-bit data block and 128-bit secret key. MITM attack was proposed by Diffie and Hellman in 1977 as a generic method to analyze symmetric-key cryptographic algorithms. SCT was proposed by Aoki and Sasaki to improve MITM attack in 2009. In this paper we show that 19-round HIGHT can be attacked with 28 bytes of memory, 28 + 2 pairs of chosen plain and cipher texts, and 2120.7 times of the encryption operation by using MITM attack with SCT.

A comparison of learning performance in two-dimensional Q-learning by the difference of Q-values alignment

In this article, we examine the learning performance of various strategies under different conditions using the Voronoi Q-value element (VQE) based on reward in a single-agent environment, and decide how to act in a certain state. In order to test our hypotheses, we performed computational experiments using several situations such as various angles of rotation of VQEs which are arranged into a lattice structure, various angles of an agent’s action rotation that has 4 actions, and a random arrangement of VQEs to correctly evaluate the optimal Q-values for state and action pairs in order to deal with continuous-valued inputs. As a result, the learning performance changes when the angle of VQEs and the angle of action are changed by a specific relative position.

Reinforcement learning using Voronoi space division

Reinforcement learning is considered an important tool for robotic learning in unknown/uncertain environments. In this article, we suggest that Voronoi space division creates a new Voronoi region which permits an arbitrary point in the plane, say a Voronoi Q-value element (VQE), and constructs a new method for space division using a Voronoi diagram in order to realize multidimensional reinforcement learning. This article shows some results for four-dimensional spaces, and the essential characteristics of VQEs in a continuous state and action are also described. The advantages of learning with a variety of VQEs are enhanced learning speed and reliability for this task.

A study of Q-learning considering negative rewards

In the reinforcement learning system, the agent obtains a positive reward, such as 1, when it achieves its goal. Positive rewards are propagated around the goal area, and the agent gradually succeeds in reaching its goal. If you want to avoid certain situations, such as dangerous places or poison, you might want to give a negative reward to the agent. However, in conventional Q-learning, negative rewards are not propagated in more than one state. In this article, we propose a new way to propagate negative rewards. This is a very simple and efficient technique for Q-learning. Finally, we show the results of computer simulations and the effectiveness of the proposed method.

One-dimensional resolution improvement of images in time series by genetic algorithm

Several frame images in a time series that can be taken by a TV camera system under specified conditions can be given improved one-dimensional resolution by solving a recurrence formula or by an inverse-filtering method. It appears that one-dimensional resolution improvement is the same as the problem of optimum search satisfying given conditions for the images in a time series. It is known that genetic algorithms are extremely effective in optimum search problems. In this paper, a genetic algorithm is investigated for one-dimensional improvement of the resolution of images in a time series. Some experimental results are presented. Also, it is shown that the proposed method is effective for improvement of the resolution of images in time series with noise.

New method for measuring the topology preservation of self-organizing feature maps

The topology preservation of self-organizing feature maps is an important property which is used in many applications. Various qualitative and quantitative approaches are known for measuring the degree of topology preservation. However, well-received measures for determining the topology preservation have not been presented yet. In this paper, we present a new method for measuring the degree of topology preservation based on the masked Delaunay triangulation. The topology preservation is completed when the masked Delaunay triangulation coincides with the output network shape. We demonstrate the usefulness of this measure for various examples of data manifolds. This method is also applied to measure the degree of topology preservation of the topology-representing network of T. Martinetz et al. (1994). In various simulations, this measure brought a reasonable degree of topology preservation.

Domain search using shrunken images for fractal image compression

In this paper, we propose a new way of limiting the number of candidates of domains by using the shrunken image for Voronoi-based fractal image compression. And we show the result of computer simulations and confirm the effects of the proposed method. The process of domain search is the most critical process of fractal image compression because it takes exorbitant time to perform it. In the process of domain search, we have to use the term of Σri, Σ di, Σri2, Σ di2 and Σridi, whereri is the sum of pixels for the ith range anddi is same one for the corresponding domain. We can calculate these terms by using cumulations for the rectangular range, but for the Voronoi range, since the shape of a range is different from each other, we can not use the cumulations for calculating these terms. Therefore, it is necessary to limit the number of candidates of domains for finding the appropriate domain in order to reduce the time of compressing image.

Two dimensional resolution improvement of frame images by genetic algorithms

It is thought that two dimensional resolution improvement of frame images is same as the problem of optimum search that satisfied a given condition to the frame imaging system. It is known that a genetic algorithm is extremely effective for optimum search problems. In this paper, a genetic algorithm is investigated for two dimensional resolution improvement of frame images in time series. Also, some experimental results are shown, and it is proved that the proposed method is effective when used with noisy images.