The article presents the use of Vector Quantised-Variational Autoencoder (VQ-VAE) neural networks to enhance the sampling process in financial statement audits. Key aspects of this research include:

• Development of a quantized representation of accounting data.
• Learning the latent factors of variation valuable for auditing.
• Potential to serve as a representative audit sample providing better insights and accuracy.

This methodology paves the way for more robust and efficient audit processes by integrating cutting-edge machine learning technologies. It showcases the transformative potential of AI in enhancing traditional auditing practices.