Centroid Estimation With Guaranteed Efficiency: A General Framework for Weakly Supervised Learning.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Weakly supervised learning (WSL) deals with incomplete, inexact, or inaccurate labels.
• Existing WSL methods often prioritize unbiasedness over statistical efficiency.
• There is a need for robust risk estimators applicable to diverse weak supervision scenarios.

Purpose of the Study:

• To propose a general framework, Centroid Estimation with Guaranteed Efficiency (CEGE), for weakly supervised learning.
• To develop an unbiased and statistically efficient risk estimator for WSL.
• To improve performance in various WSL tasks by addressing limitations of current methods.

Main Methods:

• Decomposing loss functions into label-independent and label-dependent terms.
• Constructing two auxiliary pseudo-labeled datasets with synthesized labels.
• Deriving and linearly combining unbiased centroid estimates for statistical efficiency.

Main Results:

• The proposed CEGE framework achieves both unbiasedness and high statistical efficiency in risk estimation.
• Theoretical analysis guarantees minimum variance in centroid estimation.
• Extensive experiments show CEGE outperforms existing methods on benchmark datasets for WSL.

Conclusions:

• CEGE offers a superior approach to weakly supervised learning by optimizing both unbiasedness and efficiency.
• The framework is broadly applicable to various WSL problems, including semi-supervised learning, positive-unlabeled learning, multiple instance learning, and label noise learning.
• CEGE represents a significant advancement in developing reliable machine learning models with imperfect supervision.