Learning from label proportions on high-dimensional data.

Area of Science:

• Machine Learning
• Data Science

Background:

• Learning from label proportions (LLP) is a machine learning paradigm where training data consists of bags with only class proportions available.
• Solving high-dimensional LLP problems remains a significant challenge in the field.

Purpose of the Study:

• To propose a novel algorithm, Learning from Label Proportions based on Random Forests (LLP-RF), designed to address high-dimensional LLP.
• To improve the accuracy of machine learning models in scenarios with limited label information.

Main Methods:

• Formulating a robust loss function by defining hidden class labels as random variables within bags.
• Incorporating proportion information by penalizing discrepancies between ground truth and estimated label proportions.
• Utilizing an alternating annealing method to solve the optimization model.

Main Results:

• LLP-RF demonstrates superior performance on high-dimensional datasets.
• Experimental results show competitive or better accuracies compared to existing methods.
• The algorithm effectively handles the challenges posed by high-dimensional LLP.

Conclusions:

• LLP-RF offers an effective solution for high-dimensional learning from label proportions problems.
• The proposed method provides a robust and accurate approach for scenarios with bag-level proportion data.
• This work advances the capabilities of machine learning in dealing with complex, high-dimensional data structures.