A Similarity-based Framework for Classification Task

TL;DR

This paper introduces a generalized similarity-based framework that combines similarity learning with linear models to improve multi-label classification, capturing class interdependencies and enhancing interpretability.

Contribution

It unifies similarity-based methods with generalized linear models, enabling better modeling of class relationships and robustness against noisy classes.

Findings

Effective on multi-class datasets

Improves interpretability of class contributions

Outperforms existing methods in experiments

Abstract

Similarity-based method gives rise to a new class of methods for multi-label learning and also achieves promising performance. In this paper, we generalize this method, resulting in a new framework for classification task. Specifically, we unite similarity-based learning and generalized linear models to achieve the best of both worlds. This allows us to capture interdependencies between classes and prevent from impairing performance of noisy classes. Each learned parameter of the model can reveal the contribution of one class to another, providing interpretability to some extent. Experiment results show the effectiveness of the proposed approach on multi-class and multi-label datasets