Details

Autor(en) / Beteiligte

• Darnstädt, Malte
• Ries, Christoph
• Simon, Hans Ulrich

Titel

Hierarchical design of fast Minimum Disagreement algorithms

Ist Teil von

• Theoretical computer science, 2018-03, Vol.716, p.70-88

Ort / Verlag

Elsevier B.V

Erscheinungsjahr

2018

Link zum Volltext

Quelle

EZB Electronic Journals Library

Beschreibungen/Notizen

• We compose a toolbox for the design of Minimum Disagreement algorithms. This box contains general procedures which transform (without much loss of efficiency) algorithms that are successful for some d-dimensional (geometric) concept class C into algorithms which are successful for a (d+1)-dimensional extension of C. An iterative application of these transformations has the potential of starting with a base algorithm for a trivial problem and ending up at a smart algorithm for a non-trivial problem. In order to make this working, it is essential that the algorithms are not proper, i.e., they return a hypothesis that is not necessarily a member of C. However, the “price” for using a super-class H of C is so low that the resulting time bound for achieving accuracy ε in the model of agnostic learning is significantly smaller than the time bounds achieved by the up-to-date best (proper) algorithms. We evaluate the transformation technique for d=2 on both artificial and real-life data sets and demonstrate that it provides a fast algorithm, which can successfully solve practical problems on large data sets.