Efficient variational Bayesian approximation method based on subspace optimization.

Area of Science:

• Computational statistics
• Bayesian inference
• Optimization

Background:

• Variational Bayesian approximation (VBA) is crucial for approximating intractable posterior distributions in Bayesian inference.
• Classical VBA methods exhibit slow convergence in high-dimensional scenarios, limiting their practical application.
• Addressing this limitation is essential for advancing Bayesian computational methods.

Purpose of the Study:

• To develop a more efficient variational Bayesian approximation (VBA) method.
• To enhance the convergence speed and applicability of VBA for large-dimensional problems.
• To demonstrate the method's effectiveness in image processing tasks.

Main Methods:

• The study frames the variational Bayesian problem as a functional optimization task.
• It adapts subspace optimization techniques from Hilbert spaces to function spaces for iterative solutions.
• The core innovation involves determining optimal directions at each iteration to accelerate convergence.

Main Results:

• The proposed VBA method demonstrates significantly improved computation time compared to existing state-of-the-art techniques.
• Its efficiency is validated through application to an ill-posed linear inverse problem in image processing using a total variation prior.
• Numerical examples confirm a notable reduction in computational burden.

Conclusions:

• The novel VBA method offers a more efficient approach to Bayesian inference, particularly for large-scale and computationally intensive problems.
• The adaptation of subspace optimization provides a robust framework for accelerating variational inference.
• This advancement has practical implications for fields like image processing requiring efficient approximate Bayesian solutions.