Outlier-Robust Convex Segmentation

We address the task of segmenting sequential data using convex optimization problem, which is specifically designed to work in the context of outliers in the data. We propose two algorithms for solving this problem, one exact and one a top-down hierarchical approach. Robustness to outliers is evaluated on a real-world task related to speech segmentation. Our algorithms outperform baseline segmentation algorithms.

• Type: Journal
• Year: 2015
• Where: AAAI
• Authors: Itamar Katz, Koby Crammer
• Keywords: AAAI 2015, Intelligent Agents