Learning from labeled and unlabeled data using a minimal number of queries

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Labeling data is time-consuming and expensive.
• Existing methods focus on either active learning or learning from unlabeled data.

Purpose of the Study:

• To develop an algorithm integrating active learning and learning from unlabeled data.
• To minimize labeling effort while maximizing classification accuracy.

Main Methods:

• A Genetic Algorithm (GA) refines unlabeled data class membership.
• Reverses roles of labeled and unlabeled data for implicit class assignment.
• Uses GA population to build multiple Maximum A Posteriori (MAP) classifiers for active learning selection.

Main Results:

• The integrated approach achieves accurate classification.
• Demonstrates reduced labeling effort compared to traditional methods.
• Implicit class assignment of unlabeled data is achieved.

Conclusions:

• The proposed algorithm effectively combines active learning and unsupervised learning.
• Offers a more efficient approach to data labeling and classification.
• Potential for significant cost and time savings in machine learning projects.