LCBM: A Multi-View Probabilistic Model for Multi-Label Classification.

Area of Science:

• Machine Learning
• Artificial Intelligence

Background:

• Multi-label classification is crucial in machine learning.
• Exploiting label dependencies enhances model performance.
• Probabilistic models show promise in uncovering label relationships.

Purpose of the Study:

• Propose a novel multi-view probabilistic model for multi-label classification.
• Introduce the latent conditional Bernoulli mixture (LCBM) model.
• Leverage multi-view learning to improve generalization.

Main Methods:

• Developed a generative multi-view probabilistic model (LCBM).
• Utilized a Gaussian mixture variational autoencoder (GMVAE) for posterior approximation.
• Employed a scalable stochastic training algorithm and greedy search for prediction.

Main Results:

• LCBM effectively models label dependencies within a shared latent subspace.
• The model demonstrates computational convenience through weak correlations within mixture components.
• Experimental results show superior performance over state-of-the-art methods on benchmark datasets.

Conclusions:

• The proposed LCBM model offers an effective approach to multi-label classification.
• Multi-view learning integrated with probabilistic modeling enhances performance.
• The method provides a robust and scalable solution for complex classification tasks.