arXiv Open Access 2025

Feynman-Kac Correctors in Diffusion: Annealing, Guidance, and Product of Experts

Marta Skreta Tara Akhound-Sadegh Viktor Ohanesian Roberto Bondesan Alán Aspuru-Guzik +4 lainnya
Lihat Sumber

Abstrak

While score-based generative models are the model of choice across diverse domains, there are limited tools available for controlling inference-time behavior in a principled manner, e.g. for composing multiple pretrained models. Existing classifier-free guidance methods use a simple heuristic to mix conditional and unconditional scores to approximately sample from conditional distributions. However, such methods do not approximate the intermediate distributions, necessitating additional `corrector' steps. In this work, we provide an efficient and principled method for sampling from a sequence of annealed, geometric-averaged, or product distributions derived from pretrained score-based models. We derive a weighted simulation scheme which we call Feynman-Kac Correctors (FKCs) based on the celebrated Feynman-Kac formula by carefully accounting for terms in the appropriate partial differential equations (PDEs). To simulate these PDEs, we propose Sequential Monte Carlo (SMC) resampling algorithms that leverage inference-time scaling to improve sampling quality. We empirically demonstrate the utility of our methods by proposing amortized sampling via inference-time temperature annealing, improving multi-objective molecule generation using pretrained models, and improving classifier-free guidance for text-to-image generation. Our code is available at https://github.com/martaskrt/fkc-diffusion.

Topik & Kata Kunci

Penulis (9)

M

Marta Skreta

T

Tara Akhound-Sadegh

V

Viktor Ohanesian

R

Roberto Bondesan

A

Alán Aspuru-Guzik

A

Arnaud Doucet

R

Rob Brekelmans

A

Alexander Tong

K

Kirill Neklyudov

Format Sitasi

Skreta, M., Akhound-Sadegh, T., Ohanesian, V., Bondesan, R., Aspuru-Guzik, A., Doucet, A. et al. (2025). Feynman-Kac Correctors in Diffusion: Annealing, Guidance, and Product of Experts. https://arxiv.org/abs/2503.02819

Akses Cepat

Lihat di Sumber
Informasi Jurnal
Tahun Terbit
2025
Bahasa
en
Sumber Database
arXiv
Akses
Open Access ✓