Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming

Paaßen B (2015)
Bielefeld: Bielefeld University.

Bielefelder Masterarbeit | Englisch
 
Download
OA
Abstract / Bemerkung
A core issue in machine learning is the classification of data. However, for data structures that can not easily be summarized in a feature representation, standard vectorial approaches are not suitable. An alternative approach is to represent the data not by features, but by their similarities or disimilarities to each other. In the case of sequential data, dissimilarities can be efficiently calculated by well-established alignment distances. Recently, techniques have been put forward to adapt the parameters of such alignment distances to the specific data set at hand, e.g. using gradient descent on a cost function. In this thesis we provide a comprehensive theory for gradient descent on alignment distance based on Algebraic Dynamic Programming, enabling us to adapt even sophisticated alignment distances. We focus on Affine Sequence Alignment, which we optimize by gradient descent on the Large Margin Nearest Neighbor cost function. Thereby we directly optimize the classification accuracy of the popular k-Nearest Neighbor classifier. We present a free software implementation of this theory, the TCS Alignment Toolbox, which we use for the subsequent experiments. Our experiments entail alignment distance learning on three diverse data sets (two artificial ones and one real-world example), yielding not only an increase in classification accuracy but also interpretable resulting parameter settings.
Stichworte
sequence alignment; metric learning; dissimilarity data
Jahr
2015
Seite(n)
125
Page URI
https://pub.uni-bielefeld.de/record/2736686

Zitieren

Paaßen B. Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming. Bielefeld: Bielefeld University; 2015.
Paaßen, B. (2015). Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming. Bielefeld: Bielefeld University.
Paaßen, B. (2015). Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming. Bielefeld: Bielefeld University.
Paaßen, B., 2015. Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming, Bielefeld: Bielefeld University.
B. Paaßen, Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming, Bielefeld: Bielefeld University, 2015.
Paaßen, B.: Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming. Bielefeld University, Bielefeld (2015).
Paaßen, Benjamin. Adaptive Affine Sequence Alignment Using Algebraic Dynamic Programming. Bielefeld: Bielefeld University, 2015.
Alle Dateien verfügbar unter der/den folgenden Lizenz(en):
Copyright Statement:
This Item is protected by copyright and/or related rights. [...]
Volltext(e)
Access Level
OA Open Access
Zuletzt Hochgeladen
2019-09-25T06:38:37Z
MD5 Prüfsumme
6843f0eab2e3f72e9dc09397f5c90ae3

Export

Markieren/ Markierung löschen
Markierte Publikationen

Open Data PUB

Suchen in

Google Scholar