Computing and Informatics (E-Journal - Institute of Informatics, SAS, Bratislava)
Not a member yet
1506 research outputs found
Sort by
The Algorithms for FPGA Implementation of Sparse Matrices Multiplication
Get PDFIn comparison to dense matrices multiplication, sparse matrices multiplication real performance for CPU is roughly 5--100 times lower when expressed in GFLOPs. For sparse matrices, microprocessors spend most of the time on comparing matrices indices rather than performing floating-point multiply and add operations. For 16-bit integer operations, like indices comparisons, computational power of the FPGA significantly surpasses that of CPU. Consequently, this paper presents a novel theoretical study how matrices sparsity factor influences the indices comparison to floating-point operation workload ratio. As a result, a novel FPGAs architecture for sparse matrix-matrix multiplication is presented for which indices comparison and floating-point operations are separated. We also verified our idea in practice, and the initial implementations results are very promising. To further decrease hardware resources required by the floating-point multiplier, a reduced width multiplication is proposed in the case when IEEE-754 standard compliance is not required
Parallel Computer Simulation of Fire in Road Tunnel and People Evacuation
Get PDFAdvances in CFD (Computational Fluid Dynamics) and significant increase of computational power of current computers have led to widespread use of CFD in aerodynamics, fluid dynamics, combustion engineering and other academic disciplines. One of such disciplines is computer modelling and simulation of fire in human structures. Fire is a very complicated and complex phenomenon. Fire research deals with such processes as combustion, radiation, heat transfer, turbulence, fluid dynamics, and other physical and chemical processes. Several advanced fire and smoke simulation systems have been developed to solve various aspects of fire safety in various conditions and environments. In this paper, the use of parallel version of the CFD simulator FDS (Fire Dynamics Simulator) for the simulation of fire spread and smoke development in a short road tunnel is described. In order to study the impact of the computational domain decomposition on the accuracy and reliability of simulation results, several simulations of a chosen fire scenario ran on the HP blade cluster utilizing different numbers of processors. The obtained parameters of fire and smoke were used to investigate the influence of the fire on people evacuation in the tunnel with active ventilation for a given traffic situation
Distributed Frequent Item Sets Mining over P2P Networks
Get PDFData intensive peer-to-peer (P2P) networks are becoming increasingly popular in applications like social networking, file sharing networks, etc. Data mining in such P2P environments is the new generation of advanced P2P applications. Unfortunately, most of the existing data mining algorithms do not fit well in such environments since they require data that can be accessed in its entirety. It also is not easy due to the requirements of online transactional data streams. In this paper, we have developed a local algorithm for tracing frequent item sets over a P2P network. The performance of the proposed algorithm is comparatively tested and analyzed through a series of experiments
Distributed Real-Time Computation of the Point of Gaze
Get PDFThis paper presents a minimally intrusive real-time gaze-tracking prototype to be used in several scenarios, including a laboratory stall and an in-vehicle system. The system requires specific infrared illumination to allow it to work with variable light conditions. However, it is minimally intrusive due to the use of a carefully configured switched infrared LED array. Although the perceived level of illumination generated by this array is high, it is achieved using low-emission infrared light beams. Accuracy is achieved through a precise estimate of the center of the user's pupil. To overcome inherent time restrictions while using low-cost processors, its main image-processing algorithm has been distributed over four main computing tasks. This structure not only enables good performance, but also simplifies the task of experimenting with alternative computationally-complex algorithms and with alternative tracking models based on locating both user eyes and several cameras to improve user mobility
PFS: A Productivity Forecasting System for Desktop Computers to Improve Grid Applications Performance in Enterprise Desktop Grid
Get PDFAn Enterprise Desktop Grid (EDG) is a low cost platform that gathers desktop computers spread over different institutions. This platform uses desktop computers idle time to run Grid applications. We argue that computers in these environments have a predictable productivity that affects a Grid application execution time. In this paper, we propose a system called PFS for computer productivity forecasting that improves Grid applications performance. We simulated 157.500 applications and compared the performance achieved by our proposal against two recent strategies. Our experiments show that a Grid scheduler based on PFS runs applications faster than schedulers based on other selection strategies
Flexible Fuzzy Rule Bases Evolution with Swarm Intelligence for Meta-Scheduling in Grid Computing
Get PDFFuzzy rule-based systems are expert systems whose performance is strongly related to the quality of their knowledge and the associated knowledge acquisition processes and thus, the design of effective learning techniques is considered a critical and major problem of these systems. Knowledge acquisition with a swarm intelligence approach is a recent learning strategy for the evolution of fuzzy rule bases founded on swarm intelligence showing improvement over classical knowledge acquisition strategies in fuzzy rule based systems such as Pittsburgh and Michigan approaches in terms of convergence behaviour and accuracy. In this work, a generalization of this method is proposed to allow the simultaneous consideration of diversely configured knowledge bases and this way to accelerate the learning process of the original algorithm. In order to test the suggested strategy, a problem of practical importance nowadays, the design of expert meta-schedulers systems for grid computing is considered. Simulations results show the fact that the suggested adaptation improves the functionality of knowledge acquisition with a swarm intelligence approach and it reduces computational effort; at the same time it keeps the quality of the canonical strategy
Current Sensing Completion Detection in Single-Rail Asynchronous Systems
Get PDFIn this article, an alternative approach to detecting the computation completion of combinatorial blocks in asynchronous digital systems is presented. The proposed methodology is based on well-known phenomenon that occurs in digital systems fabricated in CMOS technology. Such logic circuits exhibit significantly higher current consumption during the signal transitions than in the idle state. Duration of these current peaks correlates very well with the actual computation time of the combinatorial block. Hence, this fact can be exploited for separation of the computation activity from static state. The paper presents fundamental background of addressed alternative completion detection and its implementation in single-rail encoded asynchronous systems, the proposed current sensing circuitry, achieved simulation results as well as the comparison to the state-of-the-art methods of completion detection. The presented method promises the enhancement of the performance of an asynchronous circuit, and under certain circumstances it also reduces the silicon area requirements of the completion detection block
Statistics Oriented Preprocessing of Document Image
Get PDFOld printed documents represent an important part of our cultural heritage. Their digitalization plays an important role in creating data and metadata. The paper proposed an algorithm for estimation of the global text skew. First, document image is binarized reducing the impact of noise and uneven illumination. The binary image is statistically analyzed and processed. Accordingly, redundant data have been excluded. Furthermore, the convex hulls are established encircling each text object. They are joined establishing connected components. Then, the connected components in complementary image are enlarged with morphological dilation. At the end, the biggest connected component is extracted. Its orientation is similar to the global orientation of text document which is calculated by the moments. Efficiency and correctness of the algorithm are verified by testing on a custom dataset
A Multi-Layered Architecture for Collaborative and Decentralized Consequence Finding
Get PDFThe consequence finding problem consists in producing all the consequences of a logical theory or, depending on the application context, in a restricted subset of these consequences. When the available knowledge is naturally scattered among different sources of information, computing such consequences with respect to the global theory in a decentralized way is a challenging problem. This paper presents Somewhere2, a multilayered architecture that may be used to solve such consequence finding problems in peer-to-peer networks of collaborating entities, that may evolve over time. The general layout of this architecture is described as well as the roles of its main components. Thanks to a careful and modular design, the resulting framework is very generic. This facilitates alternative implementations of specific components as well as its extension with additional features. First experimental results are presented, illustrating the scalability and robustness of this architecture. This framework may be used as a robust building block for more advanced distributed applications, such as Peer Data Management Systems
Active Multi-Field Learning for Spam Filtering
Get PDFUbiquitous spam messages cause a serious waste of time and resources. This paper addresses the practical spam filtering problem, and proposes a universal approach to fight with various spam messages. The proposed active multi-field learning approach is based on: 1) It is cost-sensitive to obtain a label for a real-world spam filter, which suggests an active learning idea; and 2) Different messages often have a similar multi-field text structure, which suggests a multi-field learning idea. The multi-field learning framework combines multiple results predicted from field classifiers by a novel compound weight, and each field classifier calculates the arithmetical average of multiple conditional probabilities predicted from feature strings according to a data structure of string-frequency index. Comparing the current variance of field classifying results with the historical variance, the active learner evaluates the classifying confidence and regards the more uncertain message as the more informative sample for which to request a label. The experimental results show that the proposed approach can achieve the state-of-the-art performance at greatly reduced label requirements both in email spam filtering and short text spam filtering. Our active multi-field learning performance, the standard (1-ROCA) % measurement, even exceeds the full feedback performance of some advanced individual classifying algorithm