Prototype Based Recognition of Splice Sites

Hammer, Barbara; Strickert, Marc; Villmann, Thomas

doi:10.1007/10950913_2

Cited by 5 publications

(5 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The classifier size and the classification time is constant with respect to the training set size. This has been demonstrated in comparison to SVM results for a large benchmark problem in computational biology, for which GRLVQ yields solutions with the same accuracy as a comparable SVM, but of which the size is up to a factor 100 smaller depending on the training set size [18].…”

Section: Discussionmentioning

confidence: 80%

“…LVQ-type classifiers have been designed as simple and intuitive classification methods, and variants show quite good performance in practice [17,18,24]. Since they have not explicitly been designed for large margin optimization, it is interesting to see that large margin bounds can be derived also for these models.…”

Section: Discussionmentioning

confidence: 99%

“…In particular, a problem-adjustable metric in combination with LVQ can adequately deal with high-dimensional and heterogeneous data. The extension GRLVQ, which includes an adaptive diagonal metric and the cost function of GLVQ, even achieved a classification accuracy competitive to the SVM in a large scale problem from computational biology [18]. The question occurs, which improvements of LVQ can also be supported by a theoretical counterpart to accompany the empirical findings of previous work.…”

Section: Introductionmentioning

confidence: 93%

See 2 more Smart Citations

On the Generalization Ability of GRLVQ Networks

Hammer

Strickert

Villmann

2005

Neural Process Lett

Self Cite

View full text Add to dashboard Cite

We derive a generalization bound for prototype-based classifiers with adaptive metric. The bound depends on the margin of the classifier and is independent of the dimensionality of the data. It holds for classifiers based on the Euclidean metric extended by adaptive relevance terms. In particular, the result holds for relevance learning vector quantization (RLVQ) [4] and generalized relevance learning vector quantization (GRLVQ) [19].

show abstract

Section: Discussionmentioning

confidence: 80%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 93%

See 1 more Smart Citation

On the Generalization Ability of GRLVQ Networks

Hammer

Strickert

Villmann

2005

Neural Process Lett

Self Cite

View full text Add to dashboard Cite

show abstract

“…When we perform the classification based on the two features defined by the first eigendirections of Λ 1 and Λ 2 , we loose almost no performance and still achieve 94.0% ± 0.38% test accuracy. SVM results reported in the literature even lie above 96% [14,20] test accuracy. Note, however, that our classifier is extremely sparse and simple and still achieves a performance which is only slightly worse.…”

Section: (B)(c)mentioning

confidence: 84%

Adaptive Relevance Matrices in Learning Vector Quantization

2009

Self Cite

View full text Add to dashboard Cite

We propose a new matrix learning scheme to extend relevance learning vector quantization (RLVQ), an efficient prototype-based classification algorithm, towards a general adaptive metric. By introducing a full matrix of relevance factors in the distance measure, correlations between different features and their importance for the classification scheme can be taken into account and automated, general metric adaptation takes place during training. In comparison to the weighted Euclidean metric used in RLVQ and its variations, a full matrix is more powerful to represent the internal structure of the data appropriately. Large margin generalization bounds can be transfered to this case leading to bounds which are independent of the input dimensionality. This also holds for local metrics attached to each prototype which corresponds to piecewise quadratic decision boundaries. The algorithm is tested in comparison to alternative LVQ schemes using an artificial data set, a benchmark multi-class problem from the UCI repository, and a problem from bioinformatics, the recognition of splice sites for C. elegans.

show abstract

“…Also there was a recent study based on a Support Vector Machine (SVM) [45] and a method combining SVM with a Hidden Markov Model (HMM) [46]. The approach discussed in [47] combines MDD with first order Markov models, i.e. WAMs, at the leafs as compared to a simple WMM.…”

Section: Ss Sensor Design Conceptmentioning

confidence: 98%

Contemporary Progress in Gene Structure Prediction

Churbanov¹

2006

View full text Add to dashboard Cite

Despite substantial recent progress, gene structural prediction remains a challenging problem in bioinformatics. The importance of a detailed understanding of gene splicing can be underlined by noting that ~ 10-15% of human genetic diseases are caused by mutations that affect splice junctions. We briefly introduce the problem, mention the existing approaches to gene structural annotation and provide overview of current methods. In particular, this paper explains why homology-based gene structural prediction appears to be more difficult then it might seem. The problem of splice sites (SSs) sensor design is overviewed with rigorous comparison of key designs. Finally, a discussion of methods in ab initio gene structural prediction is accompanied by an extensive comparative performance study. We make certain conclusions regarding the current state of the art and try to speculate about future research directions. Applications used to evaluate performance characteristics for various gene structural prediction programs are available online at http://www. wyomingbioinformatics.org/~achurban/.

show abstract

Prototype Based Recognition of Splice Sites

Cited by 5 publications

References 53 publications

On the Generalization Ability of GRLVQ Networks

On the Generalization Ability of GRLVQ Networks

Adaptive Relevance Matrices in Learning Vector Quantization

Contemporary Progress in Gene Structure Prediction

Contact Info

Product

Resources

About