Good Similarity Learning for Structured Data
Similarity functions play an important role in the performance of many learning algorithms, thus a lot of research has gone into training them. In this paper, we focus on learning similarity functions for structured data. We propose a novel edit similarity learning approach (GESL) driven by the idea of (e,g,t)-goodness, a recent theory that bridges the gap between the properties of a similarity function and its performance in classification. We derive generalization guarantees for our method and provide experimental evidence of its practical interest.