Classification via score-based generative modelling

TL;DR

This paper explores score-based gradient learning for classification, demonstrating its ability to characterize data distributions, generate samples, and improve classification accuracy and robustness, especially in complex scenarios.

Contribution

It introduces a novel application of score matching for discriminative and generative classification, offering a flexible alternative to density-based methods.

Findings

Effective in binary classification tasks

Enhances robustness to perturbations

Works well in high-dimensional and imbalanced data scenarios

Abstract

In this work, we investigated the application of score-based gradient learning in discriminative and generative classification settings. Score function can be used to characterize data distribution as an alternative to density. It can be efficiently learned via score matching, and used to flexibly generate credible samples to enhance discriminative classification quality, to recover density and to build generative classifiers. We analysed the decision theories involving score-based representations, and performed experiments on simulated and real-world datasets, demonstrating its effectiveness in achieving and improving binary classification performance, and robustness to perturbations, particularly in high dimensions and imbalanced situations.