neural-network-approximation

This paper establishes a rigorous quantitative connection between neural network score function approximation and the resulting distribution approximation in score-based diffusion models, proving that accurate score approximation leads to close distribution approximation in KL divergence, with an explicit bound.