In face recognition, simple classifiers are frequently used. For a robust system, it is common to construct a multi-class classifier by combining the outputs of several binary classifiers; this is called output coding method. The two basic output coding methods for this purpose are known as OnePerClass (OPC) and PairWise Coupling (PWC). The performance of output coding methods depends on accuracy of base dichotomizers. Support Vector Machine (SVM) is suitable for this purpose. In this paper, we review output coding methods and introduce a new sequential fusion method using SVM as a base classifier based on OPC and PWC according to their properties. In the experiments, we compare our proposed method with others. The experimental results show that our proposed method can improve the performance significantly on the real dataset.

We propose a simple method for the chromatic dispersion measurement of optical fibers by using bi-directional modulation of a Mach-Zehnder electro-optical modulator embodied in a fiber loop mirror. The detected output of the bi-directionally modulated light, with time difference, creates fading in the RF domain. Dispersion is found by measuring the period of fading at different wavelengths.

This letter presents the advantages of a cascaded algebraic codebook structure at relatively high bit-rates. The cascaded structure that consists of two stages provides flexible pulse combinations due to an additional gain term in the second stage. The perceptual quality of the cascaded structure can be further improved by using a gain re-estimation scheme. Experiments confirm that the cascaded structure has a big advantage in terms of quality and complexity as the bit-rate becomes higher.

A Java execution environment is implemented, in which a hardware engine is operated in parallel with an embedded processor. This pair of hardware facilities together with an additional software kernel are devised for existing embedded systems, so as to execute Java applications more efficiently in such a way that 39 instructions are added to the original Java Virtual Machine to implement the software kernel. The exploration of design parameters is also attempted to attain a low hardware cost and high performance. The proposed hardware engine of a 6-stage pipeline can be integrated in a single chip using 30 k gates together with the instruction and data cache memories. The proposed approach improves the execution speed by a factor of 5 in comparison with the J2ME software implementation.

The High-Low Water Mark destage (HLWM) algorithm is widely used to enable cached RAID5 to flush dirty data from its write cache to disks due to the simplicity of its operations. It starts and stops a destaging process based on the two thresholds that are configured at the initialization time with the best knowledge of its underlying storage performance capability and its workload pattern which includes traffic intensity, access patterns, etc. However, each time the current workload varies from the original, the thresholds need to be re-configured with the changed workload. This paper proposes an efficient destage algorithm which automatically re-configures its initial thresholds according to the changed traffic intensity and access patterns, called adaptive thresholding. The core of adaptive thresholding is to define the two thresholds as the multiplication of the referenced increasing and decreasing rates of the write cache occupancy level and the time required to fill and empty the write cache. We implement the proposed algorithm upon an actual RAID system and then verify the ability of the auto-reconfiguration with synthetic workloads having a different level of traffic intensity and access patterns. Performance evaluations under well-known traced workloads reveal that the proposed algorithm reduces disk IO traffic by about 12% with a 6% increase in the overwrite ratio compared with the HLWM algorithm.

For High Performance Computing (HPC) of electromagnetic microwave simulations, the authors present concept for a microwave simulation dedicated computer, FDTD/FIT data flow machine. By constructing a dedicated computer customized to the data flow of the FDTD or FIT scheme, we can obtain maximum performance from the FDTD/FIT simulations and achieve T FLOPS performance computing by using much smaller size computer system than conventional supercomputers. In addition to the basic idea, this paper identifies with solution to some other factors which are needed to execute practical simulations (e.g., boundary condition setting, power input, simulation result data upload to PC, etc.). Moreover, the VHDL design and logical simulation of the 2D data flow machine are also shown as the first step of development of the FDTD/FIT data flow machine.

In this paper, a new grouping method for Group Technology using Self-Organizing Map (SOM) is proposed. The purpose of our study is to divide machines in a factory into any number of cells so that the machines in each cell can process a similar set of parts to increase productivity. A main feature of our method is to specify not only the number of the cells but also the maximum and minimum numbers of machines in a cell. Some experimental results show effectiveness of our proposed algorithm.

Embedded processors are used in numerous devices executing dedicated applications. This setting makes it worthwhile to optimize the processor to the application it executes, in order to increase its power-efficiency. This paper proposes to enhance direct mapped data caches with automatically tuned randomized set index functions to achieve that goal. We show how randomization functions can be automatically generated and compare them to traditional set-associative caches in terms of performance and energy consumption. A 16 kB randomized direct mapped cache consumes 22% less energy than a 2-way set-associative cache, while it is less than 3% slower. When the randomization function is made configurable (i.e., it can be adapted to the program), the additional reduction of conflicts outweighs the added complexity of the hardware, provided there is a sufficient amount of conflict misses.

Broadcast disks are suited for disseminating information to a large number of clients in mobile computing environments. In broadcast disks, the server continuously and repeatedly broadcasts all data items in the database to clients without specific requests. The clients monitor the broadcast channel and read data items as they arrive on the broadcast channel. The broadcast channel then becomes a disk from which clients can read data items. In this paper, we propose a cache conscious concurrency control (C4) algorithm to preserve the consistency of read-only client transactions, when the values of broadcast data items are updated at the server. The C4 algorithm is novel in the sense that it can reduce the response time of client transactions with minimal control information to be broadcast from the server. This is achieved by the judicious caching strategy of the client and by adjusting the currency of data items read by the client.

Proxy caches have been used for a very long time to enhance the performance of web access. Along with the recent development of CDN (Content Distribution Network), the web proxy caching has also been adopted in many main techniques. This paper presents a new viewpoint on the possible improvement to the cooperative proxy caching, which can reduce outbound traffic and therefore ideally result in better response time. We take notice to the regional total cost of cache objects for optimizing content distribution. By contrast to the regular removal policies based on single proxy server, we prefer to evaluate a retrieved web object based on the metrics gathered from multiply proxy caches regionally. We particularly introduce a concept called post-removal analysis, which is used in measuring the value of the removed objects. Finally, we use the real proxy cache Squid to implement our proposal and modify the well-known cache benchmarking tool Web Polygraph to test this cooperative prototype. The test results prove that the proposed scheme can bring noticeable improvement on the performance of proxy caching.

In this paper, we propose a machine learning-based method of multi-document summarization integrating sentence extraction with bunsetsu elimination. We employ Support Vector Machines for both of the modules used. To evaluate the effect of bunsetsu elimination, we participated in the multi-document summarization task at TSC-2 by the following two approaches: (1) sentence extraction only, and (2) sentence extraction + bunsetsu elimination. The results of subjective evaluation at TSC-2 show that both approaches are superior to the Lead-based method from the viewpoint of information coverage. In addition, we made extracts from given abstracts to quantitatively examine the effectiveness of bunsetsu elimination. The experimental results showed that our bunsetsu elimination makes summaries more informative. Moreover, we found that extraction based on SVMs trained by short extracts are better than the Lead-based method, but that SVMs trained by long extracts are not.

The efficient distribution of stored information has become a major concern in the Internet. Since the web workload characteristics show that more than 60% of network traffic is caused by image documents, how to efficiently distribute image documents from servers to end clients is an important issue. Proxy cache is an efficient solution to reduce network traffic. And it has been shown that an image caching method (Graceful Caching) based on hierarchical coding format performs better than conventional caching schemes in recent years. However, as the capacity of the cache is limited, how to efficiently allocate the cache memory to achieve a minimum expected delay time is still a problem to be resolved. This paper presents an integrated caching algorithm to deal with the above problem for image databases, web browsers, proxies and other similar applications in the Internet. By analyzing the web request distribution of the Graceful Caching, both replacing and pre-fetching algorithms are proposed. We also show that our proposal can be carried out based on information readily available in the proxy server; it flexibly adapts its parameters to the hit rates and access pattern of users' requesting documents in the Graceful Caching. Finally we verify the performance of this algorithm by simulations.

Many cooperated web cache systems and protocols have been proposed. These systems, however, require expensive resources, such as external bandwidth and CPU power or storage of a proxy, while inducing hefty administrative costs to achieve adequate client population growth. Moreover, a scalability problem in the cache server management still exists. This paper suggests peer-to-peer client-clustering. The client-cluster provides a proxy cache with backup storage which is comprised of the residual resources of the clients. We use DHT based peer-to-peer lookup protocol to manage the client-cluster. With the natural characteristics of this protocol, the client-cluster is self-organizing, fault-tolerant, well-balanced and scalable. Additionally, we propose the Backward ICP which is used to communicate between the proxy cache and the client-cluster, to reduce the overhead of the object replication and to use the resources more efficiently. We examine the performance of the client-cluster via a trace driven simulation and demonstrate effective enhancement of the proxy cache performance.

For suppressing inter symbol interference, the support vector machine mutliuser detector (SVM-MUD) was adopted as a nonlinear method in direct sequence code division multiple access (DS-CDMA) signals transmitted through multipath channels. To solve the problems of the complexity of SVM-MUD model and the number of support vectors, based on recursive least squares support vector machine (RLS-SVM) and Riemannian geometry, a new algorithm for nonlinear multiuser detector is proposed. The algorithm introduces the forgetting factor to get the support vectors at the first training samples, then, uses Riemannian geometry to train the support vectors again and gets less improved support vectors. Simulation results illustrated that the algorithm simplifies SVM-MUD model at the cost of only a little more bit error rate and decreases the computational complexity. At the same time, the algorithm has an excellent effect on suppressing multipath interference.

The wavelet transform approach is applied to the boundary element method (BEM) for solving electromagnetic scattering from multiple scatterers. A matrix equation is first obtained by the BEM where the elements of the impedance matrix are arranged as smooth as possible along its columns and rows. Consequently, the matrix is divided into several minor matrices with continuous and periodic structures along their columns and rows. Next, the matrix equation is transformed by a wavelet matrix to sparsify the impedance matrix. The wavelet matrix is constructed to consist of minor wavelet matrices and makes the minor matrices of the impedance matrix be transformed independently. This approach reduces both the computation costs of performing the wavelet transform and solving sparse linear equations if it is compared with the conventional one.

This letter presents the implementation results of an application-level cache file system, MCFS, which is specifically designed to provide efficient caching and transmission mechanisms for streaming media. The file system is built on a virtual file disk which is constructed as a single large file on a general-purpose file system. MCFS suits the access requirement of continuous media caching and provides an efficient I/O mechanism for cache servers. The experimental results show that MCFS outperforms the comparison model and provides a consistent I/O bandwidth.

We consider the classification problem as a problem of approximation of a given training set. This approximation is constructed in a multiresolution framework, and organized in a tree-structure. It allows efficient training and query, both in constant time per training point. The proposed method is efficient for low-dimensional classification and regression estimation problems with large data sets.

A set of systematic experiments on intelligent email categorization has been conducted with different machine learning algorithms applied to different parts of data in order to achieve the most correct classification. The categorization is based on not only the body but also the header of an email message. The metadata (e.g. sender name, sender organization, etc.) provide additional information that can be exploited to improve the categorization capability. Results of experiments on real email data demonstrate the feasibility of our approach to find the best learning algorithm and the metadata to be used, which is a very significant contribution in email classification. It is also shown that categorization based only on the header information is comparable or superior to that based on all the information in a message for all the learning algorithms considered.

Evaluating the generalization performance of learning machines without using additional test samples is one of the most important issues in the machine learning community. The subspace information criterion (SIC) is one of the methods for this purpose, which is shown to be an unbiased estimator of the generalization error with finite samples. Although the mean of SIC agrees with the true generalization error even in small sample cases, the scatter of SIC can be large under some severe conditions. In this paper, we therefore investigate the causes of degrading the precision of SIC, and discuss how its precision could be improved.

A simple configuration for millimeter-wave fiber-wireless transmission, with remote local-oscillator (LO) delivery from the central office, both for the uplink and for the downlink, and a simple, cost-effective, base-station solution is proposed. Under the assumption of using commercially available components and a conventional single-mode fiber (with dispersion of 17 ps/nm/km at 1.55 µm), our numerical results show that, with a laser linewidth of 150 MHz, a laser power of 0 dBm and an optical gain of only 6 dB, it is possible to transmit, without repeaters, data rates of 622 Mbit/s across about 18 km at a bit-error-rate of 10-9. By increasing the optical gain to 24 dB, the link length can be increased to approximately 67 km for a laser linewidth of 75 MHz and to 78 km for a laser linewidth 1 MHz.