Adaptive Diffusion Guidance via Stochastic Optimal Control

Iskander Azangulov, Peter Potaptchik, Qinyu Li, Eddie Aamari, George Deligiannidis, Judith Rousseau
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:4087-4095, 2026.

Abstract

Classifier-Free Guidance (CFG) is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to CFG scheduling—determining the appropriate guidance weight—are largely heuristic and lack a solid theoretical foundation. This work addresses these limitations on two fronts. First, we provide a theoretical formalization that precisely characterizes the relationship between guidance strength and classifier confidence. Second, building on this insight, we introduce a stochastic optimal control framework that casts CFG scheduling as an adaptive optimization problem. In this formulation, guidance strength is not fixed but dynamically selected based on time, the current sample, and the conditioning class, either independently or in combination. By solving the resulting control problem, we establish a principled foundation for more effective guidance in diffusion models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-azangulov26a, title = { Adaptive Diffusion Guidance via Stochastic Optimal Control }, author = {Azangulov, Iskander and Potaptchik, Peter and Li, Qinyu and Aamari, Eddie and Deligiannidis, George and Rousseau, Judith}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {4087--4095}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/azangulov26a/azangulov26a.pdf}, url = {https://proceedings.mlr.press/v300/azangulov26a.html}, abstract = { Classifier-Free Guidance (CFG) is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to CFG scheduling—determining the appropriate guidance weight—are largely heuristic and lack a solid theoretical foundation. This work addresses these limitations on two fronts. First, we provide a theoretical formalization that precisely characterizes the relationship between guidance strength and classifier confidence. Second, building on this insight, we introduce a stochastic optimal control framework that casts CFG scheduling as an adaptive optimization problem. In this formulation, guidance strength is not fixed but dynamically selected based on time, the current sample, and the conditioning class, either independently or in combination. By solving the resulting control problem, we establish a principled foundation for more effective guidance in diffusion models. } }
Endnote
%0 Conference Paper %T Adaptive Diffusion Guidance via Stochastic Optimal Control %A Iskander Azangulov %A Peter Potaptchik %A Qinyu Li %A Eddie Aamari %A George Deligiannidis %A Judith Rousseau %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-azangulov26a %I PMLR %P 4087--4095 %U https://proceedings.mlr.press/v300/azangulov26a.html %V 300 %X Classifier-Free Guidance (CFG) is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to CFG scheduling—determining the appropriate guidance weight—are largely heuristic and lack a solid theoretical foundation. This work addresses these limitations on two fronts. First, we provide a theoretical formalization that precisely characterizes the relationship between guidance strength and classifier confidence. Second, building on this insight, we introduce a stochastic optimal control framework that casts CFG scheduling as an adaptive optimization problem. In this formulation, guidance strength is not fixed but dynamically selected based on time, the current sample, and the conditioning class, either independently or in combination. By solving the resulting control problem, we establish a principled foundation for more effective guidance in diffusion models.
APA
Azangulov, I., Potaptchik, P., Li, Q., Aamari, E., Deligiannidis, G. & Rousseau, J.. (2026). Adaptive Diffusion Guidance via Stochastic Optimal Control . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:4087-4095 Available from https://proceedings.mlr.press/v300/azangulov26a.html.

Related Material