The journey of graph kernels through two decades

In the real world all events are connected. There is a hidden network of dependencies that governs behavior of natural processes. Without much argument it can be said that, of all the known data-structures, graphs are naturally suitable to model such information. But to learn to use graph data structure is a tedious job as most operations on graphs are computationally expensive, so exploring fast machine learning techniques for graph data has been an active area of research and a family of algorithms called kernel based approaches has been famous among researchers of the machine learning domain. With the help of support vector machines, kernel based methods work very well for learning with Gaussian processes. In this survey we will explore various kernels that operate on graph representations. Starting from the basics of kernel based learning we will travel through the history of graph kernels from its first appearance to discussion of current state of the art techniques in practice.     

An empirical study on improving dissimilarity-based classifications using one-shot similarity measure

This paper reports an experimental result obtained by additionally using unlabeled data together with labeled ones to improve the classification accuracy of dissimilarity-based methods, namely, dissimilarity-based classifications (DBC) [25]. In DBC, classifiers among classes are not based on the feature measurements of individual objects, but on a suitable dissimilarity measure among the objects instead. In order to measure the dissimilarity distance between pairwise objects, an approach using the one-shot similarity (OSS) [30] measuring technique instead of the Euclidean distance is investigated in this paper. In DBC using OSS, the unlabeled set can be used to extend the set of prototypes as well as to compute the OSS distance. The experimental results, obtained with artificial and real-life benchmark datasets, demonstrate that designing the classifiers in the OSS dissimilarity matrices instead of expanding the set of prototypes can further improve the classification accuracy in comparison with the traditional Euclidean approach. Moreover, the results demonstrate that the proposed setting does not work with non-Euclidean data.     

Indices for the evaluation of wildfire spread simulations using contemporaneous predictions and observations of burnt area

Methods to objectively evaluate performance are critical for model development. In contrast to recent advances in wildfire simulation, there has been limited attention to evaluating fire model performance. Information to validate fire models is typically limited, commonly to a few perimeter observations at a small number of points in time. We review metrics for comparing two burnt areas at a point in time: observed and predicted. These are compared in an idealised landscape and with a case study evaluating the performance of simulations of an Australian wildfire. We assessed: Shape Deviation Index (SDI), Jaccard's coefficient, F1, Sørensen's Similarity and Area Difference Index (ADI). For decomposing fit into error components (overprediction and underprediction) we assessed the partial indices of SDI and ADI, Precision and Recall. The various metrics were evaluated for their ability to represent error and their suitability for use in model improvement frameworks.     

Phishing attacks discovery based on HTML layout similarity

Based on the similarity of the layout structure between the phishing sites and real sites,an approach to discover phishing sites was presented.First,the tag with link attribute as a feature was extracted,and then based on the feature,the page tag sequence branch to identify website was extracted,followed by the page layout similarity-HTMLTagAntiPhish,the alignment of page tag sequence tree into the alignment of page tag sequence branches was converted,this converted two-dimention tree structure into one-dimention string structure,and finally through the substitution matrix of bioinfor-matics BLOSUM62 coding,alignment score quickly to improve the phishing sites detection efficiency was computed.A series of simulation experiments show that this approach is feasible and has higher precision and recall rates.     

SIMILARITY‐BASED RETRIEVAL WITH STRUCTURE‐SENSITIVE SPARSE BINARY DISTRIBUTED REPRESENTATIONS

We present an approach to similarity‐based retrieval from knowledge bases that takes into account both the structure and semantics of knowledge base fragments. Those fragments, or analogues, are represented as sparse binary vectors that allow a computationally efficient estimation of structural and semantic similarity by the vector dot product. We present the representation scheme and experimental results for the knowledge base that was previously used for testing of leading analogical retrieval models MAC/FAC and ARCS. The experiments show that the proposed single‐stage approach provides results compatible with or better than the results of two‐stage models MAC/FAC and ARCS in terms of recall and precision. We argue that the proposed representation scheme is useful for large‐scale knowledge bases and free‐structured database applications.     

SDF classifier revisited

This paper is addressing problems related to the construction of classifiers based on the Similarity Discriminant Function (SDF), in which the traditional vector representation of a pattern is replaced with matrix data. We introduce potential modifications of the matrix data structure and propose new variants of the SDF. The algorithms that we present were tested on images of handwritten digits and on photographs of human faces, taken from the ORL and CMU‐PIE databases. The results of experiments show that our modifications significantly improved the performance of the original SDF classifier.     

An adaptive color similarity function suitable for image segmentation and its numerical evaluation

In this article, we present an adaptive color similarity function defined in a modified hue‐saturation‐intensity color space, which can be used directly as a metric to obtain pixel‐wise segmentation of color images among other applications. The color information of every pixel is integrated as a unit by an adaptive similarity function thus avoiding color information scattering. As a direct application we present an efficient interactive, supervised color segmentation method with linear complexity respect to the number of pixels of the input image. The process has three steps: (1) Manual selection of few pixels in a sample of the color to be segmented. (2) Automatic generation of the so called color similarity image (CSI), which is a gray level image with all the gray level tonalities associated with the selected color. (3) Automatic threshold of the CSI to obtain the final segmentation. The proposed technique is direct, simple and computationally inexpensive. The evaluation of the efficiency of the color segmentation method is presented showing good performance in all cases of study. A comparative study is made between the behavior of the proposed method and two comparable segmentation techniques in color images using (1) the Euclidean metric of the a* and b* color channels rejecting L* and (2) a probabilistic approach on a* and b* in the CIE L*a*b* color space. Our testing system can be used either to explore the behavior of a similarity function (or metric) in different color spaces or to explore different metrics (or similarity functions) in the same color space. It was obtained from the results that the color parameters a* and b* are not independent of the luminance parameter L* as one might initially assume in the CIE L*a*b* color space. We show that our solution improves the quality of the proposed color segmentation technique and its quick result is significant with respect to other solutions found in the literature. The method also gives a good performance in low chromaticity, gray level and low contrast images. © 2016 Wiley Periodicals, Inc. Col Res Appl, 42, 156–172, 2017     

Using function-based patent analysis to identify potential application areas of technology for technology transfer

Under the open innovation paradigm, identification of application areas of the internally developed technologies is important to maximize the profits from them. However, many companies have failed to identify opportunity for additional applications. The fundamental reason is that companies have insufficient understanding of the potential applications of their technologies, because different industries use far different technologies and technological terminologies. However, technologies can be linked with outside industries by analyzing them from the functional perspective, because functions used in different industries are generally similar. Therefore, this article proposes a function-based patent analysis to identify the potential application areas of a technology. In general, technologies in a patent are invented for use in a specific industry, and thus a patent can be categorized into a specific industry. By identifying patents which contain a specific function, industries that use the function can be identified. Industries in which the importance of the function is relatively high can be considered as areas in which technologies performing the function have potential to be applied, and the proposed indexes evaluate the relative importance of the function within each industry. Finally, the practical usefulness of the proposed method was verified by identifying the potential areas in which vortex reduction technology can be applied.     

Evaluation of hydrophilic matrix tablets based on Carbopol® 971P and low-viscosity sodium alginate for pH-independent controlled drug release

Background: The aim of this study was to evaluate matrix tablets containing different ratios of Carbopol^® 971P (CP) to low-viscosity sodium alginate (SA) and assess their suitability for pH-independent controlled drug release. Methods: Two processing methods (physical mixing, PM and spray-drying, SD) were applied before compaction and the release from corresponding matrices was compared. The release from CP-SA PM matrices was also investigated using three model drugs (paracetamol, salicylic acid, and verapamil HCl) and two dissolution media (0.1 N HCl or phosphate buffer, pH?=?6.8), and the release rate, mechanism, and pH-dependence were characterized by fitting of Higuchi and Peppas models, and evaluation of similarity factor. Furthermore, swelling behavior of CP-SA matrix tablets was studied for evaluating its impact on drug release. Results: The processing method (SD or PM) markedly affected the drug release from CP-SA matrices. ANOVA tests showed significant effects of the CP:SA ratio and drug type on the release rate (expressed by the constant, K_H, from Higuchi model) and of the dissolution medium on the release mechanism (expressed by the exponent, n, from Peppas model). Similarity factor (f₂) indicated that the CP:SA ratios ≥?25:75 and ≥?50:50 were suitable for pH-independent release of paracetamol and salicylic acid, respectively, although for verapamil HCl, the matrix with low CP:SA ratio (0:100) showed remarkably reduced pH-dependence of release. Swelling parameters (water uptake and mass loss) were significantly changed with experimental variables (CP:SA ratio, medium, and time) and were in good correlation with drug release. Conclusion: Matrix tablets based on CP and SA form a potentially useful versatile system for pH-independent controlled drug release.