ARCHIVE Snapshot of the Rätsch lab website (Tübingen, then New York) as of February 2016: preserved, not maintained. The group today: bmi.inf.ethz.ch
Personal tools
Home → Rätsch Lab → People → Research Topics → Machine Learning → Large Scale Learning with String Kernels
 
Document Actions

Large Scale Learning with String Kernels

The goal of this project is to develop new sequence kernels that are particularly well suited for sequence analysis tasks as well as algorithms that efficiently compute the optimal SVM solution when using these kernels.

Kernel based methods such as Support Vector Machines (SVMs) have proven to be powerful for a wide range of different data analysis problems, in particular in the field of bioinformatics. They are based on the sound foundations of statistical learning and optimization theory and employ a kernel k(x, x') computing dot products in a feature space. While SVMs achieve competitive classification accuracies, training classifiers becomes prohibitively computationally expensive when using genomesized training samples. In genomic sequence analysis tasks like splice site recognition [1] or promoter identification [2], large amounts of training sequences are available and indeed needed to achieve sufficiently high classification performances.

The goal of this project is to develop new sequence kernels that are particularly well suited for sequence analysis tasks as well as algorithms that efficiently compute the optimal SVM solution when using these kernels.

People specializing in this area

Group Leader

Gunnar Rätsch, Ph.D.

Gunnar Rätsch, Ph.D.

I have worked on the development of several kernels suitable for the classification of genomic sequences, like the Weighted Degree Kernel. Moreover, together with Sören, we came up with several algorithms that allow SVM training with millions of examples which exploit special properties of the string kernels. These algorithms are implemented in the Shogun toolbox developed by Sören Sonnenburg and me.
More about Gunnar Rätsch, Ph.D....

Alumnae and Alumni

Sören Sonnenburg, Ph.D.

Sören Sonnenburg, Ph.D.

Together with Gunnar Raetsch and several other collaborators I am developing new string kernels particularly suited for genomic signal (and content) prediction. Since in many cases data is available in abundance and indeed needed to achieve state-of-the-art performance I develop efficient algorithms for their computation. In addition, I worked on fast SVM training and evaluation algorithms (e.g. the ``linadd'' algorithm with ``batch computation'' and SVM Ocas using on-demand computed string kernel features).

More about Sören Sonnenburg, Ph.D....

Publications