Learning from Label Proportions with Generative Adversarial Networks

TL;DR

This paper introduces LLP-GAN, a novel generative adversarial network-based method for learning from label proportions, enabling effective instance-level classification without distribution restrictions and demonstrating superior scalability and performance.

Contribution

The paper proposes LLP-GAN, an end-to-end GAN framework for label proportion learning that relaxes distribution assumptions and improves scalability over existing methods.

Findings

Outperforms existing LLP methods on benchmark datasets.

Provides explicit generative representation with proven global optimality.

Demonstrates scalability benefits from deep learning models.

Abstract

In this paper, we leverage generative adversarial networks (GANs) to derive an effective algorithm LLP-GAN for learning from label proportions (LLP), where only the bag-level proportional information in labels is available. Endowed with end-to-end structure, LLP-GAN performs approximation in the light of an adversarial learning mechanism, without imposing restricted assumptions on distribution. Accordingly, we can directly induce the final instance-level classifier upon the discriminator. Under mild assumptions, we give the explicit generative representation and prove the global optimality for LLP-GAN. Additionally, compared with existing methods, our work empowers LLP solver with capable scalability inheriting from deep models. Several experiments on benchmark datasets demonstrate vivid advantages of the proposed approach.