In this study, we propose a multi-view clustering approach for mining and analysing multi-view network datasets. The proposed approach is applied and evaluated on a real-world scenario for monitoring and analysing district heating (DH) network conditions and identifying substations with sub-optimal behaviour. Initially, geographical locations of the substations are used to build an approximate graph representation of the DH network. Two different analyses can further be applied in this context: step-wise and parallel-wise multi-view clustering. The step-wise analysis is meant to sequentially consider and analyse substations with respect to a few different views. At each step, a new clustering solution is built on top of the one generated by the previously considered view, which organizes the substations in a hierarchical structure that can be used for multi-view comparisons. The parallel-wise analysis on the other hand, provides the opportunity to analyse substations with regards to two different views in parallel. Such analysis is aimed to represent and identify the relationships between substations by organizing them in a bipartite graph and analysing the substations’ distribution with respect to each view. The proposed data analysis and visualization approach arms domain experts with means for analysing DH network performance. In addition, it will facilitate the identification of substations with deviating operational behaviour based on comparative analysis with their closely located neighbours.

Abstract This paper introduces a new image-based handwritten historical digit dataset named Arkiv Digital Sweden (ARDIS). The images in ARDIS dataset are extracted from 15,000 Swedish church records which were written by different priests with various handwriting styles in the nineteenth and twentieth centuries. The constructed dataset consists of three single-digit datasets and one-digit string dataset. The digit string dataset includes 10,000 samples in red–green–blue color space, whereas the other datasets contain 7600 single-digit images in different color spaces. An extensive analysis of machine learning methods on several digit datasets is carried out. Additionally, correlation between ARDIS and existing digit datasets Modified National Institute of Standards and Technology (MNIST) and US Postal Service (USPS) is investigated. Experimental results show that machine learning algorithms, including deep learning methods, provide low recognition accuracy as they face difficulties when trained on existing datasets and tested on ARDIS dataset. Accordingly, convolutional neural network trained on MNIST and USPS and tested on ARDIS provide the highest accuracies $$58.80\%$$ 58.80 % and $$35.44\%$$ 35.44 % , respectively. Consequently, the results reveal that machine learning methods trained on existing datasets can have difficulties to recognize digits effectively on our dataset which proves that ARDIS dataset has unique characteristics. This dataset is publicly available for the research community to further advance handwritten digit recognition algorithms.

We propose a higher order mining (HOM) approach for modelling, monitoring and analyzing district heating (DH) substations' operational behaviour and performance. HOM is concerned with mining over patterns rather than primary or raw data. The proposed approach uses a combination of different data analysis techniques such as sequential pattern mining, clustering analysis, consensus clustering and minimum spanning tree (MST). Initially, a substation's operational behaviour is modeled by extracting weekly patterns and performing clustering analysis. The substation's performance is monitored by assessing its modeled behaviour for every two consecutive weeks. In case some significant difference is observed, further analysis is performed by integrating the built models into a consensus clustering and applying an MST for identifying deviating behaviours. The results of the study show that our method is robust for detecting deviating and sub-optimal behaviours of DH substations. In addition, the proposed method can facilitate domain experts in the interpretation and understanding of the substations' behaviour and performance by providing different data analysis and visualization techniques.

This paper proposes a pre-training method for neural network-based character recognizers to reduce the required amount of training data, and thus the human labeling effort. The proposed method transfers knowledge about the similarities between graph representations of characters to the recognizer by training to predict the graph edit distance. We show that convolutional neural networks trained with this method outperform traditional supervised learning if only ten or less labeled images per class are available. Furthermore, we show that our approach performs up to 33% better than a graph edit distance based recognition approach, even if only one labeled image per class is available.

Energy consumption has been widely studied in the computer architecture field for decades. While the adoption of energy as a metric in machine learning is emerging, the majority of research is still primarily focused on obtaining high levels of accuracy without any computational constraint. We believe that one of the reasons for this lack of interest is due to their lack of familiarity with approaches to evaluate energy consumption. To address this challenge, we present a review of the different approaches to estimate energy consumption in general and machine learning applications in particular. Our goal is to provide useful guidelines to the machine learning community giving them the fundamental knowledge to use and build specific energy estimation methods for machine learning algorithms. We also present the latest software tools that give energy estimation values, together with two use cases that enhance the study of energy consumption in machine learning.

In the context of document image analysis, image binarization is an important preprocessing step for other document analysis algorithms, but also relevant on its own by improving the readability of images of historical documents. While historical document image binarization is challenging due to common image degradations, such as bleedthrough, faded ink or stains, achieving good binarization performance in a timely manner is a worthwhile goal to facilitate efficient information extraction from historical documents. In this paper, we propose a recurrent neural network based algorithm using Grid Long Short-Term Memory cells for image binarization, as well as a pseudo F-Measure based weighted loss function. We evaluate the binarization and execution performance of our algorithm for different choices of footprint size, scale factor and loss function. Our experiments show a significant trade-off between binarization time and quality for different footprint sizes. However, we see no statistically significant difference when using different scale factors and only limited differences for different loss functions. Lastly, we compare the binarization performance of our approach with the best performing algorithm in the 2016 handwritten document image binarization contest and show that both algorithms perform equally well.

Machine learning software accounts for a significant amount of energy consumed in data centers. These algorithms are usually optimized towards predictive performance, i.e. accuracy, and scalability. This is the case of data stream mining algorithms. Although these algorithms are adaptive to the incoming data, they have fixed parameters from the beginning of the execution. We have observed that having fixed parameters lead to unnecessary computations, thus making the algorithm energy inefficient. In this paper we present the nmin adaptation method for Hoeffding trees. This method adapts the value of the nmin parameter, which significantly affects the energy consumption of the algorithm. The method reduces unnecessary computations and memory accesses, thus reducing the energy, while the accuracy is only marginally affected. We experimentally compared VFDT (Very Fast Decision Tree, the first Hoeffding tree algorithm) and CVFDT (Concept-adapting VFDT) with the VFDT-nmin (VFDT with nmin adaptation). The results show that VFDT-nmin consumes up to 27% less energy than the standard VFDT, and up to 92% less energy than CVFDT, trading off a few percent of accuracy in a few datasets.

Outlier detection has been studied in many domains. Outliers arise due to different reasons such as mechanical issues, fraudulent behavior, and human error. In this paper, we propose an unsupervised approach for outlier detection in a sequence dataset. The proposed approach combines sequential pattern mining, cluster analysis, and a minimum spanning tree algorithm in order to identify clusters of outliers. Initially, the sequential pattern mining is used to extract frequent sequential patterns. Next, the extracted patterns are clustered into groups of similar patterns. Finally, the minimum spanning tree algorithm is used to find groups of outliers. The proposed approach has been evaluated on two different real datasets, i.e., smart meter data and video session data. The obtained results have shown that our approach can be applied to narrow down the space of events to a set of potential outliers and facilitate domain experts in further analysis and identification of system level issues.

In the context of historical document analysis, image binarization is a first important step, which separates foreground from background, despite common image degradations, such as faded ink, stains, or bleed-through. Fast binarization has great significance when analyzing vast archives of document images, since even small inefficiencies can quickly accumulate to years of wasted execution time. Therefore, efficient binarization is especially relevant to companies and government institutions, who want to analyze their large collections of document images. The main challenge with this is to speed up the execution performance without affecting the binarization performance. We modify a state-of-the-art binarization algorithm and achieve on average a 3.5 times faster execution performance by correctly mapping this algorithm to a heterogeneous platform, consisting of a CPU and a GPU. Our proposed parameter tuning algorithm additionally improves the execution time for parameter tuning by a factor of 1.7, compared to previous parameter tuning algorithms. We see that for the chosen algorithm, machine learning-based parameter tuning improves the execution performance more than heterogeneous computing, when comparing absolute execution times.

Summary Thread‐level speculation can be used to take advantage of multicore architectures for JavaScript in web applications. We extend previous studies with these main contributions; we implement thread‐level speculation in the state‐of‐the art just‐in‐time‐enabled JavaScript engine V8 and make the measurements in the Chromium web browser both from Google instead of using an interpreted JavaScript engine. We evaluate the thread‐level speculation and just‐in‐time compilation combination on 15 very popular web applications, 20 HTML5 demos from the JS1K competition, and 4 Google Maps use cases. The performance is evaluated on two, four, and eight cores. The results clearly show that it is possible to successfully combine thread‐level speculation and just‐in‐time compilation. This makes it possible to take advantage of multicore architectures for web applications while hiding the details of parallel programming from the programmer. Further, our results show an average speedup for the thread‐level speculation and just‐in‐time compilation combination by a factor of almost 3 on four cores and over 4 on eight cores, without changing any of the JavaScript source code. Copyright © 2016 John Wiley & Sons, Ltd.

GPUs in embedded platforms are reaching performance levels comparable to desktop hardware, thus it becomes interesting to apply Computer Vision techniques. We propose, implement, and evaluate a novel feature detector and descriptor combination, i.e., we combine the Harris-Hessian detector with the FREAK binary descriptor. The implementation is done in OpenCL, and we evaluate the execution time and classification performance. We compare our approach with two other methods, FAST/BRISK and ORB. Performance data is presented for the mobile device Xperia Z3 and the desktop Nvidia GTX 660. Our results indicate that the execution times on the Xperia Z3 are insufficient for real-time applications while desktop execution shows future potential. Classification performance of Harris-Hessian/FREAK indicates that the solution is sensitive to rotation, but superior in scale variant images.

This paper introduces a new metric for approximating structural instability in Bayer image data. We show that the metric can be used to identify and classify validity of color correlation in local image regions. The metric is used to improve interpolation performance of an existing state-of-the-art single pass linear demosaicing algorithm, with virtually no impact on computational GPGPU complexity and performance. Using four different image sets, the modification is shown to outperform the original method in terms of visual quality, by having an average increase in PSNR of 0.7dB in the red, 1.5dB in the green and 0.6dB in the blue channel respectively. Because of fewer high-frequency artifacts, the average output data size also decreases by 2.5%.

Data mining algorithms are usually designed to optimize a trade-off between predictive accuracy and computational efficiency. This paper introduces energy consumption and energy efficiency as important factors to consider during data mining algorithm analysis and evaluation. We extended the CRISP (Cross Industry Standard Process for Data Mining) framework to include energy consumption analysis. Based on this framework, we conducted an experiment to illustrate how energy consumption and accuracy are affected when varying the parameters of the Very Fast Decision Tree (VFDT) algorithm. The results indicate that energy consumption can be reduced by up to 92.5% (557 J) while maintaining accuracy.

Virtual platforms provide benefits to developers in terms of a more rapid development cycle since development may begin before next-generation hardware is available. However, there is a distinct lack of graphics virtualization in industry-grade virtual platforms, leading to performance issues that may reduce the benefits virtual platforms otherwise have over execution on actual hardware. This paper demonstrates graphics acceleration by the means of paravirtualizing OpenGL ES in the Wind River Simics full-system simulator. We propose a solution for paravirtualized graphics using magic instructions to share memory between target and host systems, and present an implementation utilizing this method. The study illustrates the benefits and drawbacks of paravirtualized graphics acceleration and presents a performance analysis of strengths and weaknesses compared to software rasterization. Additionally, benchmarks are devised to stress key aspects in the solution, such as communication latency and computationally intensive applications. We assess paravirtualization as a viable method to accelerate graphics in system simulators, this reduces frame times up to 34 times compared to that of software rasterization. Furthermore, magic instructions are identified as the primary bottleneck of communication latency in the implementation.

JavaScript is a sequential programming language that has a large potential for parallel execution in Web applications. Thread-level speculation can take advantage of this, but it has a large memory overhead. In this article, we evaluate the effects of adjusting various parameters for thread-level speculation. Our results clearly show that thread-level speculation is a useful technique for taking advantage of multicore architectures for JavaScript in Web applications, that nested speculation is required in thread-level speculation, and that the execution characteristics of Web applications significantly reduce the needed memory, the number of threads, and the depth of our speculation.

Machine learning algorithms are frequently applied in data mining applications. Many of the tasks in this domain concern high-dimensional data. Consequently, these tasks are often complex and computationally expensive. This paper presents a GPU-based parallel implementation of the Random Forests algorithm. In contrast to previous work, the proposed algorithm is based on the compute unified device architecture (CUDA). An experimental comparison between the CUDA-based algorithm (CudaRF), and state-of-the-art Random Forests algorithms (Fas-tRF and LibRF) shows that CudaRF outperforms both FastRF and LibRF for the studied classification task.

JavaScript has gone from being a mechanism for providing dynamic web pages to an important component of many web applications. Currently one of the most popular type of web applications is so-called social networks, e.g., Facebook, Twitter, and MySpace. However, the workload and execution behavior of JavaScript in this context have not been fully explored or understood. In this paper we present a methodology for characterizing the JavaScript execution behavior in interactive web applications using deterministic execution of use cases. Then, we apply this methodology to evaluate a set of social network applications and compare their behavior to a set of established JavaScript benchmarks. Our results confirm previous studies that the execution behavior of social networks differ from established benchmarks. In addition, we identify one novel difference not published before, i.e., the use of anonymous functions in web applications.

With an estimated installation base of around 1 billion units, the Java J2ME platform is one of the largest development targets available. For mobile devices, J2ME is often the only available environment. For the very large body of software written in C other languages, this means difficult and costly porting to another language to support J2ME devices. This paper presents the Cibyl programming environment which allows existing code written in C and other languages supported by GCC to be recompiled into Java bytecode and run with close to native Java performance on J2ME devices. Cibyl translates compiled MIPS binaries into Java bytecode. In contrast to other approaches, Cibyl supports the full C language, is based on unmodified standard tools, and does not rely on source code conversion. To achieve good performance, Cibyl employs extensions to the MIPS architecture to support low-overhead calls to native Java functionality and use knowledge of the MIPS ABI to avoid computing unused values and transfer unnecessary registers. An evaluation on multiple virtual machines shows that Cibyl achieves performance similar to native Java, with results ranging from a slowdown of around 2 to a speedup of over 9 depending on the JVM and the benchmark.

In this paper we suggest and evaluate a method for predicting fault densities in modified classes early in the development process, i.e., before the modifications are implemented. We start by establishing methods that according to literature are considered the best for predicting fault densities of modified classes. We find that these methods can not be used until the system is implemented. We suggest our own methods, which are based on the same concept as the methods suggested in the literature, with the difference that our methods are applicable before the coding has started. We evaluate our methods using three large telecommunication systems produced by Ericsson. We find that our methods provide predictions that are of similar quality to the predictions based on metrics available after the code is implemented. Our predictions are, however, available much earlier in the development process. Therefore, they enable better planning of efficient fault prevention and fault detection activities

Abstract The current trend of using multiprocessor computers for server applications requires operating system adaptations to take advantage of more powerful hardware. However, modifying large bodies of software is very costly and time consuming, and the cost of porting an operating system to a multiprocessor might not be motivated by the potential performance benefits. In this paper we present a novel method, theapplication kernel approach, for adaption of an existing uniprocessor kernel to multiprocessor hardware. Our approach considers the existing uniprocessor kernel as a ‘black box’, to which no or very small changes are made. Instead, the original kernel runs operating system services unmodified on one processor whereas the other processors execute applications on top of a small custom kernel. We have implemented the application kernel for the Linux operating system, which illustrates that the approach can be realized with fairly small resources. We also present an evaluation of the performance and complexity of our approach, where we show that it is possible to achieve good performance while at the same time keeping the implementation complexity low. Copyright © 2006 John Wiley & Sons, Ltd.

Many software systems are developed in a number of consecutive releases. Each new release does not only add new code but also modifies already existing one. In this study we have shown that the modified code can be an important source of faults. The faults are widely recognized as one of the major cost drivers in software projects. Therefore we look for methods of improving fault detection in the modified code. We suggest and evaluate a number of prediction models for increasing the efficiency of fault detection. We evaluate them against the theoretical best model, a simple model based on size, as well as against analyzing the code in a random order (not using any model). We find that using our models provides a significant improvement both over not using any model at all and using the simple model based on the class size. The gain offered by the models corresponds to 30% to 60% of the theoretical maximum.

The architecture of a large complex software system, i.e., the division of the system into components and modules, is crucial since it often affects and limits the quality attributes of the system, e.g., performance and maintainability. In this paper we evaluate three software components for intra- and inter-process communication in a distributed real-time system, i.e., an automated guided vehicle system. We evaluate three quality attributes: performance, maintainability, and portability. The performance and maintainability are evaluated quantitatively using prototype-based evaluation, while the portability is evaluated qualitatively. Our findings indicate that it might be possible to use one third-party component for both intra- and inter-process communication, thus replacing two in-house developed components.

The ongoing transition from uniprocessor to multiprocessor computers requires operating system support. There is a large body of specialized operating systems which require porting in order to work on multiprocessors. In this paper we describe the design and implementation of a multiprocessor port of a cluster operating system kernel. The multiprocessor support is implemented with a giant locking scheme, which allowed us to get an initial working version with only minor changes to the original code. We also discuss performance and implementation experiences.

Summary form only given. The current trend of using multiprocessor computers for server applications requires operating system adoptions for high performance. However, modifying large bodies of software is very costly and time-consuming, and the cost of porting an operating system to a multiprocessor might not be motivated by the potential performance benefits. We present a novel method, the application kernel approach, for adaption of an existing uniprocessor kernel to SMP hardware. Our approach considers the existing kernel as a "black box", to which no or small changes are made. Instead, the original kernel runs OS-services unmodified on one processor whereas the other processors execute applications on top of a small custom kernel. A prototype implementation illustrates that the approach can be realized with fairly small resources. We also present an initial performance evaluation where we show that the performance is good for user-bound applications.

The cost, complexity, and inflexibility of hardware-based directory protocols motivate us to study the performance implications of protocols that emulate directory management using software handlers executed on the compute processors. An important performance limitation of such software-only protocols is that software latency associated with directory management ends up on the critical memory access path for read miss transactions. We propose five strategies that support efficient data transfers in hardware whereas directory management is handled at a slower pace in the background by software handlers. Simulations show that this approach can remove the directory-management latency from the memory access path. Whereas the directory is managed in software, the hardware mechanisms must access the memory state in order to enable data transfers at a high speed. Overall, our strategies reach between 60% and 86% of the hardware-based protocol performance.