Diffusion Model이 sample quality 측면에 있어서 높은 fidelity를 자랑한다면, density estimation은 어떨까? log-likelihood를 최소화하도록 Diffusion Model에 약간의 변형을 줘 본, 아래 2개의 논문을 리뷰해보자.

  1. Song et al. (2021). Maximum likelihood training of score-based diffusion models.
  2. Kingma et al. (2021). Variational diffusion models.

1. Likelihood Weighting

2. VDM

Summary

Reference

Websites

[사이트 출처 1]

Papers

[1] Song, Y., Durkan, C., Murray, I., & Ermon, S. (2021). Maximum likelihood training of score-based diffusion models. Advances in Neural Information Processing Systems, 34, 1415-1428.

[2] Kingma, D., Salimans, T., Poole, B., & Ho, J. (2021). Variational diffusion models. Advances in neural information processing systems, 34, 21696-21707.

댓글남기기