Solving Bayesian inverse problems with diffusion priors and off-policy RL
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910872556797952 |
|---|---|
| author | Scimeca, Luca Venkatraman, Siddarth Jain, Moksh Kim, Minsu Sendera, Marcin Hasan, Mohsin Rowe, Luke Mittal, Sarthak Lemos, Pablo Bengio, Emmanuel Adam, Alexandre Rector-Brooks, Jarrid Hezaveh, Yashar Perreault-Levasseur, Laurence Bengio, Yoshua Berseth, Glen Malkin, Nikolay |
| author_facet | Scimeca, Luca Venkatraman, Siddarth Jain, Moksh Kim, Minsu Sendera, Marcin Hasan, Mohsin Rowe, Luke Mittal, Sarthak Lemos, Pablo Bengio, Emmanuel Adam, Alexandre Rector-Brooks, Jarrid Hezaveh, Yashar Perreault-Levasseur, Laurence Bengio, Yoshua Berseth, Glen Malkin, Nikolay |
| contents | This paper presents a practical application of Relative Trajectory Balance (RTB), a recently introduced off-policy reinforcement learning (RL) objective that can asymptotically solve Bayesian inverse problems optimally. We extend the original work by using RTB to train conditional diffusion model posteriors from pretrained unconditional priors for challenging linear and non-linear inverse problems in vision, and science. We use the objective alongside techniques such as off-policy backtracking exploration to improve training. Importantly, our results show that existing training-free diffusion posterior methods struggle to perform effective posterior inference in latent space due to inherent biases. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_09746 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Solving Bayesian inverse problems with diffusion priors and off-policy RL Scimeca, Luca Venkatraman, Siddarth Jain, Moksh Kim, Minsu Sendera, Marcin Hasan, Mohsin Rowe, Luke Mittal, Sarthak Lemos, Pablo Bengio, Emmanuel Adam, Alexandre Rector-Brooks, Jarrid Hezaveh, Yashar Perreault-Levasseur, Laurence Bengio, Yoshua Berseth, Glen Malkin, Nikolay Machine Learning Artificial Intelligence This paper presents a practical application of Relative Trajectory Balance (RTB), a recently introduced off-policy reinforcement learning (RL) objective that can asymptotically solve Bayesian inverse problems optimally. We extend the original work by using RTB to train conditional diffusion model posteriors from pretrained unconditional priors for challenging linear and non-linear inverse problems in vision, and science. We use the objective alongside techniques such as off-policy backtracking exploration to improve training. Importantly, our results show that existing training-free diffusion posterior methods struggle to perform effective posterior inference in latent space due to inherent biases. |
| title | Solving Bayesian inverse problems with diffusion priors and off-policy RL |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2503.09746 |