Active semi-supervised learning for biological data classification.

Area of Science:

• Machine Learning
• Bioinformatics
• Data Science

Background:

• Addressing the challenge of large unlabeled datasets versus scarce labeled data in machine learning.
• Highlighting the trade-off between expert annotation costs and the need for substantial labeled data for robust classifiers.
• Recognizing the limitations of existing methods in meeting interactive response time requirements for real-world applications.

Purpose of the Study:

• To propose an effective and efficient active semi-supervised learning framework.
• To introduce a novel active learning method prioritizing informative samples.
• To evaluate the framework's performance in biological datasets.

Main Methods:

• Integrating active learning for sample selection with semi-supervised learning for label propagation.
• Developing a new active learning strategy based on diversity and uncertainty criteria.
• Conducting extensive experiments on ALL-AML, Escherichia coli, and PlantLeaves II datasets.
• Comparing the proposed framework against state-of-the-art supervised and semi-supervised classifiers.

Main Results:

• The proposed framework achieves higher accuracies with fewer annotated samples and reduced training time.
• The diversity and uncertainty-based selection criterion effectively prioritizes informative boundary samples.
• Achieved performance gains of up to 20% compared to other learning techniques.
• Demonstrated a superior trade-off between accuracy and computational time compared to active supervised learning.

Conclusions:

• The active semi-semi-supervised learning framework offers a more efficient approach to classifier training.
• The method is particularly beneficial in scenarios with limited labeled data and time constraints.
• The proposed active learning strategy enhances the selection of critical data points for improved model performance.