Hierarchical Design of Fast Minimum Disagreement Algorithms

Abstract. 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 ...

• Type: Journal
• Year: 2015
• Where: ALT
• Authors: Malte Darnstädt, Christoph Ries, Hans Ulrich Simon
• Keywords: ALT 2015, Machine Learning