Exponential Families for Conditional Random Fields

TL;DR

This paper introduces a novel framework for conditional random fields using reproducing kernel Hilbert spaces, connecting them to Gaussian Process classification, and proposes efficient optimization techniques leveraging kernel decompositions.

Contribution

It develops a kernel-based formulation for conditional random fields, establishes theoretical decomposition results, and introduces efficient optimization methods.

Findings

Kernel-based conditional random fields are connected to Gaussian Process classification.

Decomposition results facilitate efficient optimization.

Stationarity can be exploited to improve computational efficiency.

Abstract

In this paper we de ne conditional random elds in reproducing kernel Hilbert spaces and show connections to Gaussian Process classi cation. More speci cally, we prove decomposition results for undirected graphical models and we give constructions for kernels. Finally we present e cient means of solving the optimization problem using reduced rank decompositions and we show how stationarity can be exploited e ciently in the optimization process.