A reinforcement learning approach to online clustering.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Online clustering algorithms are crucial for analyzing streaming data.
• Competitive learning methods offer a framework for clustering but can be limited.
• Integrating reinforcement learning provides a novel approach to enhance clustering strategies.

Purpose of the Study:

• To propose a general technique for embedding online clustering algorithms within a reinforcement learning framework.
• To introduce the reinforcement guided competitive learning (RGCL) algorithm.
• To enhance clustering capabilities and performance through sustained exploration.

Main Methods:

• Viewing clustering systems as reinforcement learning systems.
• Developing the reinforcement guided competitive learning (RGCL) algorithm as a reinforcement-based adaptation of learning vector quantization (LVQ).
• Implementing extensions of RGCL and LVQ with sustained exploration properties.

Main Results:

• The proposed RGCL algorithm demonstrates enhanced clustering capabilities.
• Extensions incorporating sustained exploration significantly improve algorithm performance.
• Experimental tests on well-known datasets validate the effectiveness of the proposed methods.

Conclusions:

• The integration of reinforcement learning offers a powerful paradigm for advancing online clustering.
• RGCL and its extensions provide effective solutions for complex clustering tasks.
• Sustained exploration is a key factor in improving the performance of competitive learning algorithms.