This paper studies a cascade system of dynamic binary neural networks. The system is characterized by signum activation function, ternary connection parameters, and integer threshold parameters. As a fundamental learning problem, we consider storage and stabilization of one desired binary periodic orbit that corresponds to control signals of switching circuits. For the storage, we present a simple method based on the correlation learning. For the stabilization, we present a sparsification method based on the mutation operation in the genetic algorithm. Using the Gray-code-based return map, the storage and stability can be investigated. Performing numerical experiments, effectiveness of the learning method is confirmed.

抄録全体を表示

Machine learning is used in various fields and demand for implementations is increasing. Within machine learning, a Random Forest is a multi-class classifier with high-performance classification, achieved using bagging and feature selection, and is capable of high-speed training and classification. However, as a type of ensemble learning, Random Forest determines classifications using the majority of multiple trees; so many decision trees must be built. Performance increases with the number of decision trees, requiring memory, and decreases if the number of decision trees is decreased. Because of this, the algorithm is not well suited to implementation on small-scale hardware as an embedded system. As such, we have proposed Boosted Random Forest, which introduces a boosting algorithm into the Random Forest learning method to produce high-performance decision trees that are smaller. When evaluated using databases from the UCI Machine learning Repository, Boosted Random Forest achieved performance as good or better than ordinary Random Forest, while able to reduce memory use by 47%. Thus, it is suitable for implementing Random Forests on embedded hardware with limited memory.

抄録全体を表示

Much effort has been applied to research on object detection by statistical learning methods in recent years, and the results of that work are expected to find use in fields such as ITS and security. Up to now, the research has included optimization of computational algorithms for real-time processing on hardware such as GPU's and FPGAs. Such optimization most often works only with particular parameters, which often forfeits the flexibility that comes with dynamic changing of the target object. We propose a hardware architecture for faster detection and flexible target reconfiguration while maintaining detection accuracy. Tests confirm operation in a practical time when implemented in an FPGA board.

抄録全体を表示

High-availability seamless redundancy (HSR) is a fault-tolerant protocol for Ethernet networks that provides two frame copies for each frame sent. Each copy is forwarded on a separate physical path. HSR is a potential candidate for several fault-tolerant Ethernet applications including smart grid communications. However, one of the drawbacks of the HSR protocol is that it generates and circulates unnecessary frames within connected rings regardless of the presence of a destination node in the ring. This downside will degrade network performance and may deplete network resources. Previously, we proposed a simple but efficient approach to solving the above problem, namely, port locking (PL), which is based on the media access control address. The PL approach enables the network to learn the locations of the source and destination nodes gradually for each connection pair without using network control frames; the PL then prunes all the rings that do not contain the destination node by locking the corresponding ring's entrance ports at its QuadBox node. In this paper, we present an enhanced port-locking (EPL) approach that increases the number of pruned unused HSR rings. The analysis and corresponding simulation results show that the network traffic volume is significantly reduced for a large-sized HSR connected-rings network and consequently, network performance is greatly improved compared to the standard HSR protocol, and even PL.

抄録全体を表示

In this paper, we propose an utilization-aware hybrid beacon scheduling method for a large-scale IEEE 802.15.4 cluster-tree ZigBee network. The proposed method aims to enhance schedulability of a target network by better utilizing transmission medium, while avoiding inter-cluster collisions at the same time. To achieve this goal, the proposed scheduling method partially allows beacon overlaps, if appropriate. In particular, this paper answers for the following questions: 1) on which condition clusters can send overlapped beacons, 2) how to select clusters to overlap with minimizing utilization, and 3) how to adjust beacon parameters for grouped clusters. Also, we quantitatively evaluate the proposed method compared to previous works — i.e., non-beacon scheduling and a serialized beacon scheduling algorithm — from several aspects including total duty cycles, packet drop rate, and end-to-end delay.

抄録全体を表示

A key issue in Peer-to-Peer (P2P) live streaming systems is that several participant peers tend to leave within a short time period. For example, such a phenomenon is common at the half time of football games and at the end of the performance of famous artists. Such selfish behavior of the participants causes several problems in P2P networks such as the disconnection of the overlay, the departure of backup peers and the occurrence of cyclic reference to backup peers. In this paper, we propose several techniques for tree-structured P2P live streaming systems to enhance their resilience to the simultaneous departure of some participants. As the baseline of the discussion, we will focus on mTreebone which is a typical churn-resilient P2P live streaming system based on the notion of peer stability. The performance of the proposed techniques is evaluated by simulation. The simulation result indicates that even under high churn rates, the proposed techniques significantly reduce the number of attempts needed to connect to backup peers and the recovery time after a fail.

抄録全体を表示

In this paper, a new approach is proposed for extracting the spatio-temporal patterns from a location-based social networking system (SNS) such as Foursquare. The proposed approach consists of the following procedures. First, the spatio-temporal behaviors of users in SNS are approximated as a probabilistic distribution by using a diffusion-type formula. Since the SNS datasets generally consist of sparse check-in's of users at some time points and locations, it is difficult to investigate the spatio-temporal patterns on a wide range of time and space scales. The proposed method can estimate such wide range patterns by smoothing the sparse datasets by a diffusion-type formula. It is crucial in this method to estimate robustly the scale parameter by giving a prior generative model on check-in's of users. The robust estimation enables the method to extract appropriate patterns even in small local areas. Next, the covariance matrix among the time points is calculated from the estimated distribution. Then, the principal eigenfunctions are approximately extracted as the spatio-temporal patterns by principal component analysis (PCA). The distribution is a mixture of various patterns, some of which are regular ones with a periodic cycle and some of which are irregular ones corresponding to transient events. Though it is generally difficult to separate such complicated mixtures, the experiments on an actual Foursquare dataset showed that the proposed method can extract many plausible and interesting spatio-temporal patterns.

抄録全体を表示

To overcome the target-aspect sensitivity in radar high resolution range profile (HRRP) recognition, a novel method called Improved Kernel Distance Fuzzy C-means Clustering Method (IKDFCM) is proposed in this paper, which introduces kernel function into fuzzy c-means clustering and relaxes the constraint in the membership matrix. The new method finds the underlying geometric structure information hiding in HRRP target and uses it to overcome the HRRP target-aspect sensitivity. The relaxing of constraint in the membership matrix improves anti-noise performance and robustness of the algorithm. Finally, experiments on three kinds of ground HRRP target under different SNRs and four UCI datasets demonstrate the proposed method not only has better recognition accuracy but also more robust than the other three comparison methods.

抄録全体を表示

This paper proposes a novel approach to generate stereo video in which the zoom magnification is not constant. Although this has been achieved mechanically in a conventional way, it is necessary for this approach to develop a mechanically complex system for each stereo camera system. Instead of a mechanical solution, we employ an approach from the software side: by using a pair of zoomed and non-zoomed video, a part of the non-zoomed video image is cut out and super-resolved for generating stereo video without a special hardware. To achieve this, (1) the zoom magnification parameter is automatically determined by using distributions of intensities, and (2) the cutout image is super-resolved by using optically zoomed images as exemplars. The effectiveness of the proposed method is quantitatively and qualitatively validated through experiments.

抄録全体を表示

In this letter we present an imperceptible and robust watermarking algorithm that uses a cryptographic hash function in the authentication application of digital medical imaging. In the proposed scheme we combine discrete Fourier transform (DFT) and local image masking to detect the watermark after a geometrical distortion and improve its imperceptibility. The image quality is measured by metrics currently used in digital image processing, such as VSNR, SSIM and PSNR.

抄録全体を表示

Even though cloud users want to keep their data on clouds secure, it is not easy to protect the data because cloud administrators could be malicious and hypervisor could be compromised. To solve this problem, hardware-based memory isolation schemes have been proposed. However, the data in virtual storage are not protected by the memory isolation schemes, and thus, a guest OS should encrypt the data. In this paper, we address the problems of the previous schemes and propose a hardware-based storage isolation scheme. The proposed scheme enables to protect user data securely and to achieve performance improvement.

抄録全体を表示

Traditional sparse representation-based methods for human action recognition usually pool over the entire video to form the final feature representation, neglecting any spatio-temporal information of features. To employ spatio-temporal information, we present a novel histogram representation obtained by statistics on temporal changes of sparse coding coefficients frame by frame in the spatial pyramids constructed from videos. The histograms are further fed into a support vector machine with a spatial pyramid matching kernel for final action classification. We validate our method on two benchmarks, KTH and UCF Sports, and experiment results show the effectiveness of our method in human action recognition.

抄録全体を表示

A salient feature extraction algorithm is proposed to improve the recognition rate of the speech emotion. Firstly, the spectrogram of the emotional speech is calculated. Secondly, imitating the selective attention mechanism, the color, direction and brightness map of the spectrogram is computed. Each map is normalized and down-sampled to form the low resolution feature matrix. Then, each feature matrix is converted to the row vector and the principal component analysis (PCA) is used to reduce features redundancy to make the subsequent classification algorithm more practical. Finally, the speech emotion is classified with the support vector machine. Compared with the tradition features, the improved recognition rate reaches 15%.

抄録全体を表示

Segmenting foreground objects in unconstrained dynamic scenes still remains a difficult problem. We present a novel unsupervised segmentation approach that allows robust object segmentation of dynamic scenes with large displacements. To make this possible, we project motion based foreground region hypotheses generated via standard optical flow onto visual saliency regions. The motion hypotheses correspond to inside seeds mapping of the motion boundary. For visual saliency, we generalize the image signature method from images to videos to delineate saliency mapping of object proposals. The mapping of image signatures estimated in Discrete Cosine Transform (DCT) domain favor stand-out regions in the human visual system. We leverage a Markov random field built on superpixels to impose both spatial and temporal consistence constraints on the motion-saliency combined segments. Projecting salient regions via an image signature with inside mapping seeds facilitates segmenting ambiguous objects from unconstrained dynamic scenes in presence of large displacements. We demonstrate the performance on fourteen challenging unconstrained dynamic scenes, compare our method with two state-of-the-art unsupervised video segmentation algorithms, and provide quantitative and qualitative performance comparisons.

抄録全体を表示

To eliminate casting shadows of moving objects, which cause difficulties in vision applications, a novel method is proposed based on Visual background extractor by altering its updating mechanism using relevant spatiotemporal information. An adaptive threshold and a spatial adjustment are also employed. Experiments on typical surveillance scenes validate this scheme.

抄録全体を表示