Adversarial Label Learning

TL;DR

This paper introduces adversarial label learning, a weakly supervised method that trains classifiers without labels by adversarially selecting labels within constraints, outperforming existing methods.

Contribution

It presents a novel adversarial approach to weakly supervised learning that minimizes an error bound, effectively handling bias and dependencies in weak supervision.

Findings

Successfully trains classifiers without labels on real datasets.

Outperforms existing weakly supervised learning methods.

Uses primal-dual subgradient descent for optimization.

Abstract

We consider the task of training classifiers without labels. We propose a weakly supervised method---adversarial label learning---that trains classifiers to perform well against an adversary that chooses labels for training data. The weak supervision constrains what labels the adversary can choose. The method therefore minimizes an upper bound of the classifier's error rate using projected primal-dual subgradient descent. Minimizing this bound protects against bias and dependencies in the weak supervision. Experiments on three real datasets show that our method can train without labels and outperforms other approaches for weakly supervised learning.