Boost Picking: A Universal Method on Converting Supervised Classification to Semi-supervised Classification

TL;DR

This paper introduces Boost Picking, a universal semi-supervised learning method that effectively trains classifiers using mostly unlabeled data by leveraging two weak classifiers, with theoretical guarantees and practical improvements.

Contribution

The paper presents Boost Picking, a novel semi-supervised method that converts supervised classification into semi-supervised learning with theoretical proof of effectiveness.

Findings

Boost Picking can train models with mostly unlabeled data under certain conditions.

Boost Picking achieves comparable performance to fully supervised models.

Test along with Training (TawT) enhances model generalization.

Abstract

This paper proposes a universal method, Boost Picking, to train supervised classification models mainly by un-labeled data. Boost Picking only adopts two weak classifiers to estimate and correct the error. It is theoretically proved that Boost Picking could train a supervised model mainly by un-labeled data as effectively as the same model trained by 100% labeled data, only if recalls of the two weak classifiers are all greater than zero and the sum of precisions is greater than one. Based on Boost Picking, we present "Test along with Training (TawT)" to improve the generalization of supervised models. Both Boost Picking and TawT are successfully tested in varied little data sets.