On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data
This addresses a fundamental limitation in machine learning for scenarios where labeled data is scarce, offering a novel solution for training classifiers with minimal supervision.
The paper tackles the problem of training binary classifiers using only unlabeled data, proving that unbiased risk estimation is impossible with a single set of unlabeled data but becomes feasible with two sets having different class priors. It proposes an ERM-based method that trains deep models and outperforms state-of-the-art approaches in experiments.
Empirical risk minimization (ERM), with proper loss function and regularization, is the common practice of supervised classification. In this paper, we study training arbitrary (from linear to deep) binary classifier from only unlabeled (U) data by ERM. We prove that it is impossible to estimate the risk of an arbitrary binary classifier in an unbiased manner given a single set of U data, but it becomes possible given two sets of U data with different class priors. These two facts answer a fundamental question---what the minimal supervision is for training any binary classifier from only U data. Following these findings, we propose an ERM-based learning method from two sets of U data, and then prove it is consistent. Experiments demonstrate the proposed method could train deep models and outperform state-of-the-art methods for learning from two sets of U data.