Search-based Structured Prediction

TL;DR

Searn is a versatile meta-algorithm that transforms complex structured prediction tasks into simpler classification problems, allowing flexible loss functions and feature classes with theoretical performance guarantees.

Contribution

Introduces Searn, a novel search-based meta-algorithm that unifies structured prediction and classification with broad applicability and theoretical assurances.

Findings

Searn can handle any loss function and feature class.

It provides strong theoretical guarantees for structured prediction performance.

Applicable to diverse domains like NLP, speech, biology, and vision.

Abstract

We present Searn, an algorithm for integrating search and learning to solve complex structured prediction problems such as those that occur in natural language, speech, computational biology, and vision. Searn is a meta-algorithm that transforms these complex problems into simple classification problems to which any binary classifier may be applied. Unlike current algorithms for structured learning that require decomposition of both the loss function and the feature functions over the predicted structure, Searn is able to learn prediction functions for any loss function and any class of features. Moreover, Searn comes with a strong, natural theoretical guarantee: good performance on the derived classification problems implies good performance on the structured prediction problem.