Search

Fault tolerance has become a major concern in exascale computing, especially for the large scale CPU/GPU heterogeneous clusters. The performance/cost benefit of GPU based system is subject to their abilities to provide high reliability, availability, and serviceability. The traditional CPU-based checkpoint technologies have been deployed on the GPU platform but all of them treat the GPU as a second class controllable and shared entity. As existing GPU checkpoint/restart implementations do not support checkpointing the internal GPU status, the codes running on GPU (kernel) can not be checked/restored just like the CPU codes, all the checkpoint operation is done outside the kernel. In this paper, we propose a hybrid checkpoint technology, HKC (Hybrid Kernel Checkpoint). HKC combines the PTX stub inject technology and dynamic library hijack mechanism, to save/store the internal state of a GPU kernel. Our evaluation shows that HKC increases the system reliability of CPU/GPU hybrid system with a very reasonable cost, and show more resilience than other checkpoint scheme. [ABSTRACT FROM AUTHOR]

A multiple kernel fusion method combining two multiresolution histogram face descriptors is proposed to create a powerful representation method for face recognition. The multi resolution histogram descriptors are based on local binary patterns and local phase coding to achieve invariance to various types of image degradation. The multi-kernel fusion is based on the computationally efficient spectral regression KDA. The proposed face recognition method is evaluated on FRGC 2.0 database yielding very impressive results. [ABSTRACT FROM AUTHOR]

Separation kernels are key components in embedded applications. Their small size and widespread use in high-integrity environments make them good targets for formal modelling and verification. We summarise results from the mechanisation of a separation kernel scheduler using the Z/Eves theorem prover. We concentrate on key data structures to model scheduler operations. The results are part of an experiment in a Grand Challenge in software verification, as part of a pilot project in verified OS kernels. The project aims at creating a mechanised formal model of kernel components that gets refined to code. This provides a set of reusable components, proof strategies, and general lemmas. Important findings about properties and requirements are also discussed. [ABSTRACT FROM AUTHOR]

High-resolution nuclear magnetic resonance (NMR) spectra contain important biomarkers that have potentials for early diagnosis of disease and subsequent monitoring of its progression. Traditional features extraction and analysis methods have been carried out in the original frequency spectrum domain. In this study, we conduct feature selection based on a complex wavelet transform by making use of its energy shift-insensitive property in a multiresolution signal decomposition. A false discovery rate based multiple testing procedure is employed to identify important metabolite features. Furthermore, a novel kernel-induced random forest algorithm is used for the classification of NMR spectra based on the selected features. Our experiments with real NMR spectra showed that the proposed method leads to significant reduction in misclassification rate. [ABSTRACT FROM AUTHOR]

Gene time series microarray experiments have been widely used to unravel the genetic machinery of biological process. However, most temporal gene expression data often contain noise, missing data points, and non-uniformly sampled time points, which will make the traditional analyzing methods to be unapplicable. One main approach to solve this problem is to reconstruct each gene expression profile as a continuous function of time. Then the continuous representation enables us to overcome problems related to sampling rate differences and missing values. In this paper, we introduce a novel reconstruction approach based on the support vector regression method. The proposed approach utilizes a framelet based kernel, which has the ability to approximate functions with multiscale structure and can reduce the influence of noise in data. To compensate the inadequate information from noisy and short gene expression data, we use its correlated genes as the test set to choose the optimal parameters. We show that this treatment can help to avoid over-fitting. Experimental results demonstrate that our method can improve the reconstruction accuracy. [ABSTRACT FROM AUTHOR]

This paper is concerned with relevance ranking in search, particularly that using term dependency information. It proposes a novel and unified approach to relevance ranking using the kernel technique in statistical learning. In the approach, the general ranking model is defined as a kernel function of query and document representations. A number of kernel functions are proposed as specific ranking models in the paper, including BM25 Kernel, LMIR Kernel, and KL Divergence Kernel. The approach has the following advantages. (1) The (general) model can effectively represent different types of term dependency information and thus can achieve high performance. (2) The model has strong connections with existing models such as BM25 and LMIR. (3) It has solid theoretical background. (4) The model can be efficiently computed. Experimental results on web search dataset and TREC datasets show that the proposed kernel approach outperforms MRF and other baseline methods for relevance ranking. [ABSTRACT FROM AUTHOR]

In multi-label classification problems, some samples belong to multiple classes simultaneously and thus the classes are not mutually exclusive. How to characterize this kind of correlations between labels has been a key issue for designing a new multi-label classification approach. In this paper, we define two objective functions, i.e., the number of relevant and irrelevant label pairs which are ranked incorrectly, and the model regularization term, which depict the correlations between labels and the model complexity respectively. Then a new kernel machine for multi-label classification is constructed using two-objective minimization and solved by fast and elitist multi-objective genetic algorithm, i.e., NSGA-II. Experiments on the benchmark data set Yeast illustrate that our multi-label method is a competitive candidate for multi-label classification, compared with several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Performance Macromodeling facilitates accelerated analog circuit synthesis. It usually consist of two steps: feasibility design space identification and performance macromodels generation. A feasibility design space is defined as a multidimensional space in which every design satisfies all the design constraints. The minimum set of constraints is the one that ensures the correct functionality of the given circuit topology. Performance macromodels are only constructed and thereby valid in the functionally correct design space. Support vector machines (SVMs) are used as classifier to identify the feasible design space of analog circuits. A kernel is an integral part of the SVM and contributes in obtaining an optimized and accurate classifier. The most commonly used kernels are Radial Basis Function (RBF), polynomial, spline, multilayer perceptron. In this paper, some new kernels and some other kernels composed through modifications on the some of the standard kernels, are explored. The classifiers using these kernel functions have been tested on different analog circuits in order to identify the feasible design space. HSPICE has been used for generation of learning data. Least Square SVM toolbox interfaced with MATLAB was used for classification. We found that use of modified kernels improves classification accuracy as well as shortens classifier generation time. [ABSTRACT FROM AUTHOR]

We describe an application of the novel Support Vector Machined Kernel (SVM΄ed Kernel) to the Recognition of hand-drawn shapes. The SVM΄ed kernel function is itself a support vector machine classifier that is learned statistically from data using an automatically generated training set. We show that the new kernel manages to change the classical methodology of defining a feature vector for each pattern. One will only need to define features representing the similarity between two patterns allowing many details to be captured in a concise way. In addition, we illustrate that features describing a single pattern could also be used in this new framework. In this paper we show how the SVM΄ed Kernel is defined and trained for the multiclass shape recognition problem. Simulation results show that the SVM΄ed Kernel outperforms all other classical kernels and is more robust to hard test sets. [ABSTRACT FROM AUTHOR]

Commutative semifields in odd characteristic can be equivalently described by planar functions (also known as PN functions). We describe a method to construct a semifield which is canonically associated to a planar function and use it to derive information on the nuclei directly from the planar function. This is used to determine the nuclei of families of new commutative semifields of dimensions 9 and 12 in arbitrary odd characteristic. [ABSTRACT FROM AUTHOR]

We present a new method for 2D barcode extraction based on object keypoint selection and line detection for the orientation finding of the barcode. In the interest point detection stage, we apply SURF for fast and robust keypoint detection. Next, we perform a non-linear SVM classification to detect the barcode location. We perform local threshold based on the barcode location in the image before applying a morphological operation to segment the barcode from the background information. Finally, we apply edge detection and Hough Transform to extract information of the barcode content. [ABSTRACT FROM AUTHOR]

A kernel in a directed graph D(V,E) is a set S of vertices of D such that no two vertices in S are adjacent and for every vertex u in ]> there is a vertex v in S , such that (u,v) is an arc of D. The problem of existence of a kernel is NP-complete for a general digraph. In this paper we introduce the strong kernel problem of an undirected graph G and solve it in polynomial time for circulant graphs. [ABSTRACT FROM AUTHOR]

In this survey, we look at kernelization: algorithms that transform in polynomial time an input to a problem to an equivalent input, whose size is bounded by a function of a parameter. Several results of recent research on kernelization are mentioned. This survey looks at some recent results where a general technique shows the existence of kernelization algorithms for large classes of problems, in particular for planar graphs and generalizations of planar graphs, and recent lower bound techniques that give evidence that certain types of kernelization algorithms do not exist. [ABSTRACT FROM AUTHOR]

We study the parameterized complexity of the pseudo-achromatic number problem: Given an undirected graph and a parameter k, determine if the graph can be partitioned into k groups such that every two groups are connected by at least one edge. This problem has been extensively studied in graph theory and combinatorial optimization. We show that the problem has a kernel of at most (k − 2)(k + 1) vertices that is constructable in time ]> , where n and m are the number of vertices and edges, respectively, in the graph, and k is the parameter. This directly implies that the problem is fixed-parameter tractable. We also study generalizations of the problem and show that they are parameterized intractable. [ABSTRACT FROM AUTHOR]

The One-Class Support Vector Machine (OC-SVM) is an unsupervised learning algorithm, identifying unusual or outlying points (outliers) from a given dataset. In OC-SVM, it is required to set the regularization hyperparameter and kernel hyperparameter in order to obtain a good estimate. Generally, cross-validation is often used which requires multiple runs with different hyperparameters, making it very slow. Recently, the solution path algorithm becomes popular. It can obtain every solution for all hyperparameters in a single run rather than re-solve the optimization problem multiple times. Generalizing from previous algorithms for solution path in SVMs, this paper proposes a complete set of solution path algorithms for OC-SVM, including a ν-path algorithm and a kernel-path algorithm. In the kernel-path algorithm, a new method is proposed to avoid the failure of algorithm due to indefinite matrix . Using those algorithms, we can obtain the optimum hyperparameters by computing an entire path solution with the computational cost O(n2 + cnm3) on ν-path algorithm or O(cn3 + cnm3) on kernel-path algorithm (c: constant, n: the number of sample, m: the number of sample which on the margin). [ABSTRACT FROM AUTHOR]

A code ]> is a quaternary linear code if ]> is a subgroup of ℤ . In this paper, the rank and dimension of the kernel for ℤ4-linear codes, which are the corresponding binary codes of quaternary linear codes, are studied. The possible values of these two parameters for ℤ4-linear codes, giving lower and upper bounds, are established. For each possible rank r between these bounds, the construction of a ℤ4-linear code with rank r is given. Equivalently, for each possible dimension of the kernel k, the construction of a ℤ4-linear code with dimension of the kernel k is given. [ABSTRACT FROM AUTHOR]

This paper presents an application program interface (API) that enables applications running on top of a POSIX operating system to use application-defined scheduling algorithms in a way compatible with the scheduling model of the Ada 95 Real-Time Systems Annex. Several application-defined schedulers, implemented as special user tasks, can coexist in the system in a predictable way. This API is currently implemented on our operating system MaRTE OS. [ABSTRACT FROM AUTHOR]