Learning High-Dimensional Evolving Data Streams With Limited Labels.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Streaming data presents challenges like concept drift, label scarcity, and high dimensionality.
• Existing methods often require extensive labeled data, which is impractical for real-world streaming applications.
• Addressing the curse of dimensionality and limited labels is crucial for effective stream learning.

Purpose of the Study:

• To propose a new semisupervised learning technique for streaming data.
• To overcome the limitations of high dimensionality and label scarcity in data stream learning.
• To develop a method that is more practical for real-world streaming scenarios.

Main Methods:

• A denoising autoencoder is used to reduce high-dimensional feature spaces.
• A cluster-and-label technique with dynamic microclusters is employed to minimize reliance on true labels.
• A disagreement-based learning method is utilized to handle concept drift.

Main Results:

• The proposed semisupervised learning technique demonstrates superior performance.
• The method effectively handles high dimensionality and label scarcity in streaming data.
• Experimental results show significant improvements over state-of-the-art methods on real-world datasets.

Conclusions:

• The developed technique offers a practical and effective solution for semisupervised learning on data streams.
• The combination of feature reduction, dynamic clustering, and disagreement-based learning addresses key challenges in stream learning.
• This approach advances the field of learning from data streams with limited supervision.