Er K. Seewald, Florian Kleedorfer, Ge Money Bank · 2005
Abstract. The Support Vector Machine is a powerful learning algorithm which is known to work well for a variety of learning tasks. Kernels for complex data structures such as strings, lists, trees and general graphs have been developed. However, some of these kernels are of high computational complexity and are therefore not widely used. Even relatively efficient kernels may take a lot of patience to run on real-life data. We advance the state of the art for a specific kernel defined on a complex data structure, the Subsequence String Kernel (SSK) due to [9], in two ways: Firstly, by introducing an approximation technique for the SSK called Lambda Pruning (SSK-LP), which is able to reduce memory consumption and runtime by up to several orders of magnitude with little loss in accuracy. Secondly, by creating an average case time and worst case space complexity model for both SSK and SSK-LP. Estimations of runtime and memory requirements for a specific learning task can be directly computed from average string length and desired parameter settings. This combined approach has allowed us to run learning tasks which would be infeasible with the non-approximated SSK. Memory consumption for SSK-LP is constant and does not depend on string length. Models based on SSK and SSK-LP perform similarily for a set of real-life learning tasks. 1
Cited by 0
Kubernát, Tomáš · Brno University of Technology Digital Library (Brno University of Technology) · 2010
Cílem práce bylo implementovat čtyři řetězcové funkce do knihovny libSVM . Za pomoci této knihovny a výše zmíněných řetězcových funkcí poté provést sérii testování s různými hodnotami parametrů ovlivňujících výpočet samotných řetězcových funkcí. Pomocí experimentů byla porovnána rychlost a úspěšnost klasifikace mojí implementace řetězcových funkcí v knihovně libSVM s implementací řetězcových funkcí v programu kernels . V práci jsou také popsány průběhy všech testování i s naměřenými hodnotami a grafy pro grafické znázornění výsledků.
Cited by 0
Guangyue Xu, Weimin Zheng, Haiping Wu et al. · 2010 Seventh International Conference on Fuzzy Systems and Knowledge Discovery · 2010
Online databases maintain a collection of structured domain-specific documents dynamically generated in response to users' queries instead of being accessed by static URLs. Categorizing deep webs according to their object domains is a critical step to integrate such sources. While existing methods focus on supervised or post-query methodologies, we propose a more practical pre-query algorithm operating in an unsupervised manner. Given the domain number, our two phase approach firstly investigates the hidden domain distribution for each query form using topic models and each query form's object domain can be identified preliminarily. In this phase, we construct our training set composing the query forms deemed to have already been categorized correctly, and beside, the deep webs needed to be reclassified are also selected in this phase. In the second phase, we train a classifier with String Kernel methods to reclassify the uncertain deep webs to improve the overall performance. The advantage of our algorithm over previous ones is that we capture the semantic structure for each query form. Based on the two phase architecture, our framework works in an unsupervised manner and achieves satisfactory results. Experiments on the TEL-8 dataset from the UIUC Web integration repository <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> show the effectiveness and efficiency of our algorithm.
Cited by 1
Ritambhara Singh, Jack Lanchantin, Gabriel Robins et al. · IEEE/ACM Transactions on Computational Biology and Bioinformatics · 2016
Through sequence-based classification, this paper tries to accurately predict the DNA binding sites of transcription factors (TFs) in an unannotated cellular context. Related methods in the literature fail to perform such predictions accurately, since they do not consider sample distribution shift of sequence segments from an annotated (source) context to an unannotated (target) context. We, therefore, propose a method called "Transfer String Kernel" (TSK) that achieves improved prediction of transcription factor binding site (TFBS) using knowledge transfer via cross-context sample adaptation. TSK maps sequence segments to a high-dimensional feature space using a discriminative mismatch string kernel framework. In this high-dimensional space, labeled examples of the source context are re-weighted so that the revised sample distribution matches the target context more closely. We have experimentally verified TSK for TFBS identifications on 14 different TFs under a cross-organism setting. We find that TSK consistently outperforms the state-of-the-art TFBS tools, especially when working with TFs whose binding sequences are not conserved across contexts. We also demonstrate the generalizability of TSK by showing its cutting-edge performance on a different set of cross-context tasks for the MHC peptide binding predictions.
Cited by 4Open access
Singh, Ritambhara, Lanchantin, Jack, Robins, Gabriel et al. · arXiv (Cornell University) · 2016
Through sequence-based classification, this paper tries to accurately predict the DNA binding sites of transcription factors (TFs) in an unannotated cellular context. Related methods in the literature fail to perform such predictions accurately, since they do not consider sample distribution shift of sequence segments from an annotated (source) context to an unannotated (target) context. We, therefore, propose a method called "Transfer String Kernel" (TSK) that achieves improved prediction of transcription factor binding site (TFBS) using knowledge transfer via cross-context sample adaptation. TSK maps sequence segments to a high-dimensional feature space using a discriminative mismatch string kernel framework. In this high-dimensional space, labeled examples of the source context are re-weighted so that the revised sample distribution matches the target context more closely. We have experimentally verified TSK for TFBS identifications on fourteen different TFs under a cross-organism setting. We find that TSK consistently outperforms the state-of the-art TFBS tools, especially when working with TFs whose binding sequences are not conserved across contexts. We also demonstrate the generalizability of TSK by showing its cutting-edge performance on a different set of cross-context tasks for the MHC peptide binding predictions.
Cited by 0Open access
Tim vor der Brück, Yu-Fang Helena Wang · Frontiers in artificial intelligence and applications · 2012
Synonyms are a highly relevant information source for natural language processing. Automatic synonym extraction methods have in common that they are either applied on the surface representation of the text or on a syntactical structure derived from it. In this paper, however, we present a semantic synonym extraction approach that operates directly on semantic networks (SNs), which were derived from text by a deep syntactico-semantic analysis.
Cited by 0Open access
Reiji Teramoto, Mikio Aoki, Tōru Kimura et al. · FEBS Letters · 2005
Small interfering RNAs (siRNAs) are becoming widely used for sequence-specific gene silencing in mammalian cells, but designing an effective siRNA is still a challenging task. In this study, we developed an algorithm for predicting siRNA functionality by using generalized string kernel (GSK) combined with support vector machine (SVM). With GSK, siRNA sequences were represented as vectors in a multi-dimensional feature space according to the numbers of subsequences in each siRNA, and subsequently classified with SVM into effective or ineffective siRNAs. We applied this algorithm to published siRNAs, and could classify effective and ineffective siRNAs with 90.6%, 86.2% accuracy, respectively.
Cited by 64
Hao Yu, Xiaoyan Zhu, Minlie Huang · 2006
Peptides that bind to major histocompatibility complex (MHC) molecules can be presented to T-cell receptor and trigger immune response. Identification of specific binding peptides is critical for immunology research and vaccine design. A variety of methods such as HMM and ANN have been applied to predict peptides that can bind to MHC class I molecules and therefore the number of candidate binders for experimental assay can be largely reduced. However, it is a more complex process to predict peptides that bind to MHC class II molecules. In this paper, we present a SVM-based method for the prediction of MHC class II binding peptides by using string kernel. The proposed method adopts a special string kernel to compute the similarity between biological sequences with various lengths and experimental results show that our method outperforms other reported approaches. The proposed method does not require that sequences be aligned to the same length, and hence is easily employed for other prediction tasks
Cited by 3
Hugh Shanahan, Marko Babić, Piotr Byzia · 2011
Cited by 0
Saghi Nojoomi, Koehl, Patrice · Figshare · 2017
Method: Optimization of an amino acid substitution matrix. Comprehensive description of the method for optimizing the substitution matrix given as input to WSeqKernel. (PDF 809 kb)
Cited by 0Open access
Assfalg, J., Borgwardt, K., Kriegel, H. · MPG.PuRe (Max Planck Society) · 2006
Classification of 3D objects remains an important task in many areas of data management such as engineering, medicine or biology. As a common preprocessing step in current approaches to classification of voxelized 3D objects, voxel representations are transformed into a feature vector description.In this article, we introduce an approach of transforming 3D objects into feature strings which represent the distribution of voxels over the voxel grid. Attractively, this feature string extraction can be performed in linear runtime with respect to the number of voxels. We define a similarity measure on these feature strings that counts common k-mers in two input strings, which is referred to as the spectrum kernel in the field of kernel methods. We prove that on our feature strings, this similarity measure can be computed in time linear to the number of different characters in these strings. This linear runtime behavior makes our kernel attractive even for large datasets that occur in many application domains. Furthermore, we explain that our similarity measure induces a metric which allows to combine it with an M-tree for handling of large volumes of data. Classification experiments on two published benchmark datasets show that our novel approach is competitive with the best state-of-the-art methods for 3D object classification.
Cited by 1
Saghi Nojoomi, Patrice Koehl · Figshare · 2017
Method: Optimization of an amino acid substitution matrix. Comprehensive description of the method for optimizing the substitution matrix given as input to WSeqKernel. (PDF 809 kb)
Cited by 0Open access
Hancheol Park, Gahgene Gweon, Ho‐Jin Choi et al. · Waseda University Repository (Waseda University) · 2014
Paraphrase generation is widely used for various natural language processing (NLP) applications such as question answering, multi-document summarization, and machine translation. In this study, we identify the problems occurring in the process of applying existing probabilistic model-based methods to agglutinative languages, and provide solutions by reflecting the inherent characteristics of agglutinative languages. More specifically, we propose and evaluate a sentential paraphrase generation (SPG) method for the Korean language using Support Vector Machines (SVM) with a string kernel. The quality of generated paraphrases is evaluated using three criteria: (1) meaning preservation, (2) grammaticality, and (3) equivalence. Our experiment shows that the proposed method outperformed a probabilistic model-based method by 12%, 16%, and 17%, respectively, with respect to the three criteria.
Cited by 4
Ritambhara Singh, Arshdeep Sekhon, Jack Lanchantin et al. · bioRxiv (Cold Spring Harbor Laboratory) · 2018
Abstract String Kernel (SK) techniques, especially those using gapped k -mers as features (gk), have obtained great success in classifying sequences like DNA, protein, and text. However, the state-of-the-art gk-SK runs extremely slow when we increase the dictionary size ( Σ ) or allow more mismatches ( M ). This is because current gk-SK uses a trie-based algorithm to calculate co-occurrence of mismatched substrings resulting in a time cost proportional to O ( Σ M ). We propose a fast algorithm for calculating Ga pped k -mer K ernel using Co unting (GaKCo). GaKCo uses associative arrays to calculate the co-occurrence of substrings using cumulative counting. This algorithm is fast, scalable to larger Σ and M , and naturally parallelizable. We provide a rigorous asymptotic analysis that compares GaKCo with the state-of-the-art gk-SK. Theoretically, the time cost of GaKCo is independent of the Σ M term that slows down the trie-based approach. Experimentally, we observe that GaKCo achieves the same accuracy as the state-of-the-art and outperforms its speed by factors of 2, 100, and 4, on classifying sequences of DNA (5 datasets), protein (12 datasets), and character-based English text (2 datasets). 1
Cited by 4Open access
Saghi Nojoomi, Koehl, Patrice · Figshare · 2017
Method: Optimization of an amino acid substitution matrix. Comprehensive description of the method for optimizing the substitution matrix given as input to WSeqKernel. (PDF 809 kb)
Cited by 0Open access
Johannes Aßfalg, Karsten M. Borgwardt, Hans-peter Kriegel · 2006
Classification of 3D objects remains an important task in many areas of data management such as engineering, medicine or biology. As a common preprocessing step in current approaches to classification of voxelized 3D objects, voxel representations are transformed into a feature vector description. In this article, we introduce an approach of transforming 3D objects into feature strings which represent the distribution of voxels over the voxel grid. Attractively, this feature string extraction can be performed in linear runtime with respect to the number of voxels. We define a similarity measure on these feature strings that counts common k-mers in two input strings, which is referred to as the spectrum kernel in the field of kernel methods. We prove that on our feature strings, this similarity measure can be computed in time linear to the number of different characters in these strings. This linear runtime behavior makes our kernel attractive even for large datasets that occur in many application domains. Furthermore, we explain that our similarity measure induces a metric which allows to combine it with an M-tree for handling of large volumes of data. Classification experiments on two published benchmark datasets show that our novel approach is competitive with the best state-of-the-art methods for 3D object classification.
Cited by 0
Lei Li · 2009
This paper presents a method that translates false instance in grammatical database to kernel matrix through String Kernel,and uses KPCA to extract feature of the translated kernel matrix.We can separate the original false instance database into many small characteristic tables according to these extracted features,and design a classified machine by constructing false instance characteristic table.A new sentence is distributed to some characteristic table for matching of false instance through this classification machine.For characteristic table is much little than original false instance database,the running speed is enhanced very much without decreasing the accuracy of grammatical check.By compared with grammar inspection function of word,the new system demonstrates more quick speed while keeping the accuracy of grammatical check.
Cited by 0
Saghi Nojoomi, Patrice Koehl · Figshare · 2017
Method: Optimization of an amino acid substitution matrix. Comprehensive description of the method for optimizing the substitution matrix given as input to WSeqKernel. (PDF 809 kb)
Cited by 0Open access
Vanschoenwinkel, Feng Liu, Manderick · 2005
This paper extends the idea of weighted distance functions to kernels and support vector machines. Here, we focus on applications that rely on sliding a window over a sequence of string data. For this type of problems it is argued that a symbolic, context-based representation of the data should be preferred over a continuous, real format as this is a much more intuitive setting for working with (weighted) distance functions. It is shown how a weighted string distance can be decomposed and subsequently used in different kernel functions and how these kernel functions correspond to inner products between real vectors. As a case-study named entity recognition is used with information gain ratio as a weighting scheme.
Cited by 8
Microsoft Research · Bulletin of Miscellaneous Information (Royal Gardens Kew) · 2004
Kernel-based learning methods revolve around the notion of a Gram matrix between data points. These square, symmetric, positive semi-definite matrices can informally be regarded as encoding pairwise similarity between all of the objects in a data-set. In this talk I present an algorithm for manipulating the diagonal entries of a kernel matrix using semi-definite programming. Kernel matrix diagonal dominance reduction attempts to deal with the problem of learning with almost orthogonal features, a phenomenon commonplace in kernel matrices derived from string kernels or Gaussian kernels with small width parameter. I will show how this task can be formulated as a semi-definite programming optimization problem that can be solved with readily available optimizers. Theoretically I will provide an analysis using Rademacher based bounds to provide an alternative motivation for the 1-norm SVM motivated from kernel diagonal reduction. Joint work with Thore Graepel (Microsoft Research, Cambridge UK) and John Shawe-Taylor (University of Southampton, UK) ©2004 Microsoft Corporation. All rights reserved.
Cited by 0