Deterministic Annealing for Multiple-Instance Learning
Peter Gehler and Olivier Chapelle
In: AISTATS 2007, 21-24 Mar 2007, San Juan, Puerto Rico.

This is the latest version of this eprint.

Abstract

In this paper we apply deterministic annealing to different SVM formulations of the multipleinstance learning (MIL) problem. These nonconvex problems are typically solved using heuristic methods. The replacement of the integer programming formulation of the SVM formulations for MIL opens up the possibility to extend the scope of SVMs for this problem and we present a way to extend the objective function in order to incorporate more prior knowledge about the data at hand. In the experimental section we evaluate our method and compare it to the existing formulations showing that a better objective function not always translates into a better test error for datasets of little ambiguity.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.

Available Versions of this Item

• Deterministic Annealing for Multiple-Instance Learning (deposited 23 February 2007) [Currently Displayed]

[Edit]