writers!
| reviewer | homepage |
|---|---|
| ssbae | https://seongsubae.info/ |
| mjlee | https://mjbooo.github.io/ |
| ebcho | https://eunbyeol-cho.github.io/ |
paper list!
| # | Date | Reviewer | Title |
|---|---|---|---|
| 1 | 22-07-29 | mjlee | Classifier-Free Diffusion Guidance (Ho et al., 2021) |
| 2 | 22-08-05 | ssbae | Diffusion-LM Improves Controllable Text Generation (XL Li et al., 2022) |
| 3 | 22-09-20 | mjlee | [Structured Denoising Diffusion Models in Discrete State-Spaces (Ho et al., 2021)] |
후보들
Improved Denoising Diffusion Probabilistic Models (Nichol et al., 2021)
- DDPM에 몇 가지 modification을 더해서 높은 log-likelihood, 안정적인 학습 양상을 얻었음을 주장하는 논문
- Introducing Improved DDPM which can be stably trained and achieve high log-likelihood by adding few modifications to DDPM)
- https://arxiv.org/abs/2102.09672
Classifier-Free Diffusion Guidance (Ho et al., 2021)
- Diffusion Models Beat GANs on Image Synthesis에서 수행한 classifier-guidance를 pre-trained classifier 없이도 (classifier-free) 수행할 수 있음을 보인 논문
- Introducing classifier-free guidance which can perform the functionality of classifier-guidance without the pre-trained classifier, necessitated in previous paper ‘Diffusion Models Beat GANs on Image Synthesis’
- https://openreview.net/forum?id=qw8AKxfYbI
Structured Denoising Diffusion Models in Discrete State-Spaces (Ho et al., 2021)
- 기존 Multinomial diffusion model을 generalize시킴으로써 discrete state space에서 더 잘 작동하게 된 diffusion model (D3PM)을 선보인 논문
- Introducing Discrete Denoising Diffusion Probabilistic Model (D3PM) which works better in discrete state space as it generalizes the existing multinomial diffusion model.
- https://openreview.net/forum?id=h7-XixPCAL
Imagen
- TBC..
Glide (?)
- TBC..
High-Resolution Image Synthesis with Latent Diffusion Models: Diffusion for latent vectors
- TBC..
Diffusion Autoencoders: Toward a Meaningful and Decodable Representation
- TBC..