Value Gradient Sampler: Learning Invariant Value Functions for Equivariant Diffusion Sampling

Himchan Hwang, Hyeokju Jeong, Dong Kyu Shin, Che-Sang Park, Sehee Kweon, Sangwoong Yoon, Frank C. Park
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:3979-3987, 2026.

Abstract

We propose the Value Gradient Sampler (VGS), a diffusion sampler parameterized by value functions. VGS generates samples from an unnormalized target density (i.e., energy) by evolving randomly initialized particles along the gradient of the value function. In many sampling problems where the target density exhibits invariant symmetries, value functions provide a novel approach to leveraging invariant networks for sampling by inducing an equivariant gradient flow, without requiring more complex equivariant networks. The value networks are trained via temporal difference learning, which supports off-policy training and other established reinforcement learning (RL) techniques. By combining advanced RL methods with efficient invariant networks, VGS achieves both the highest sample quality and the fastest sampling speed among our baselines on the 55-particle Lennard-Jones system.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-hwang26a, title = { Value Gradient Sampler: Learning Invariant Value Functions for Equivariant Diffusion Sampling }, author = {Hwang, Himchan and Jeong, Hyeokju and Shin, Dong Kyu and Park, Che-Sang and Kweon, Sehee and Yoon, Sangwoong and Park, Frank C.}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {3979--3987}, 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/hwang26a/hwang26a.pdf}, url = {https://proceedings.mlr.press/v300/hwang26a.html}, abstract = { We propose the Value Gradient Sampler (VGS), a diffusion sampler parameterized by value functions. VGS generates samples from an unnormalized target density (i.e., energy) by evolving randomly initialized particles along the gradient of the value function. In many sampling problems where the target density exhibits invariant symmetries, value functions provide a novel approach to leveraging invariant networks for sampling by inducing an equivariant gradient flow, without requiring more complex equivariant networks. The value networks are trained via temporal difference learning, which supports off-policy training and other established reinforcement learning (RL) techniques. By combining advanced RL methods with efficient invariant networks, VGS achieves both the highest sample quality and the fastest sampling speed among our baselines on the 55-particle Lennard-Jones system. } }
Endnote
%0 Conference Paper %T Value Gradient Sampler: Learning Invariant Value Functions for Equivariant Diffusion Sampling %A Himchan Hwang %A Hyeokju Jeong %A Dong Kyu Shin %A Che-Sang Park %A Sehee Kweon %A Sangwoong Yoon %A Frank C. Park %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-hwang26a %I PMLR %P 3979--3987 %U https://proceedings.mlr.press/v300/hwang26a.html %V 300 %X We propose the Value Gradient Sampler (VGS), a diffusion sampler parameterized by value functions. VGS generates samples from an unnormalized target density (i.e., energy) by evolving randomly initialized particles along the gradient of the value function. In many sampling problems where the target density exhibits invariant symmetries, value functions provide a novel approach to leveraging invariant networks for sampling by inducing an equivariant gradient flow, without requiring more complex equivariant networks. The value networks are trained via temporal difference learning, which supports off-policy training and other established reinforcement learning (RL) techniques. By combining advanced RL methods with efficient invariant networks, VGS achieves both the highest sample quality and the fastest sampling speed among our baselines on the 55-particle Lennard-Jones system.
APA
Hwang, H., Jeong, H., Shin, D.K., Park, C., Kweon, S., Yoon, S. & Park, F.C.. (2026). Value Gradient Sampler: Learning Invariant Value Functions for Equivariant Diffusion Sampling . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:3979-3987 Available from https://proceedings.mlr.press/v300/hwang26a.html.

Related Material