Normalizing Flows with Iterative Denoising

Tianrong Chen, Jiatao Gu, David Berthelot, Joshua M. Susskind, Shuangfei Zhai
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:14367-14376, 2026.

Abstract

Normalizing Flows (NFs) are a classical family of likelihood based methods that have received revived attention. Recent efforts such as TARFlow have shown that NFs are capable to achieving promising performance on image modeling tasks, making them promising alternatives to other methods such as diffusion models. In this work, we further advance the state of Normalizing Flow generative models by introducing iterative TARFlow (iTARFlow). Unlike diffusion models, iTARFlow maintains a fully end-to-end, likelihood-based objective during training. During sampling, it performs autoregressive generation followed by an iterative denoising procedure inspired by diffusion-style methods. Through extensive experiments, We show that iTARFlow achieves competitive performance across ImageNet resolutions of 64, 128, and 256 pixels, demonstrating its potential as a strong generative model and advances the frontier of Normalizing Flows. In addition, we analyze the characteristic artifacts produced by iTARFlow, offering insights that may shed the light for future improvements.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26aj, title = {Normalizing Flows with Iterative Denoising}, author = {Chen, Tianrong and Gu, Jiatao and Berthelot, David and Susskind, Joshua M. and Zhai, Shuangfei}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {14367--14376}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/chen26aj/chen26aj.pdf}, url = {https://proceedings.mlr.press/v306/chen26aj.html}, abstract = {Normalizing Flows (NFs) are a classical family of likelihood based methods that have received revived attention. Recent efforts such as TARFlow have shown that NFs are capable to achieving promising performance on image modeling tasks, making them promising alternatives to other methods such as diffusion models. In this work, we further advance the state of Normalizing Flow generative models by introducing iterative TARFlow (iTARFlow). Unlike diffusion models, iTARFlow maintains a fully end-to-end, likelihood-based objective during training. During sampling, it performs autoregressive generation followed by an iterative denoising procedure inspired by diffusion-style methods. Through extensive experiments, We show that iTARFlow achieves competitive performance across ImageNet resolutions of 64, 128, and 256 pixels, demonstrating its potential as a strong generative model and advances the frontier of Normalizing Flows. In addition, we analyze the characteristic artifacts produced by iTARFlow, offering insights that may shed the light for future improvements.} }
Endnote
%0 Conference Paper %T Normalizing Flows with Iterative Denoising %A Tianrong Chen %A Jiatao Gu %A David Berthelot %A Joshua M. Susskind %A Shuangfei Zhai %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-chen26aj %I PMLR %P 14367--14376 %U https://proceedings.mlr.press/v306/chen26aj.html %V 306 %X Normalizing Flows (NFs) are a classical family of likelihood based methods that have received revived attention. Recent efforts such as TARFlow have shown that NFs are capable to achieving promising performance on image modeling tasks, making them promising alternatives to other methods such as diffusion models. In this work, we further advance the state of Normalizing Flow generative models by introducing iterative TARFlow (iTARFlow). Unlike diffusion models, iTARFlow maintains a fully end-to-end, likelihood-based objective during training. During sampling, it performs autoregressive generation followed by an iterative denoising procedure inspired by diffusion-style methods. Through extensive experiments, We show that iTARFlow achieves competitive performance across ImageNet resolutions of 64, 128, and 256 pixels, demonstrating its potential as a strong generative model and advances the frontier of Normalizing Flows. In addition, we analyze the characteristic artifacts produced by iTARFlow, offering insights that may shed the light for future improvements.
APA
Chen, T., Gu, J., Berthelot, D., Susskind, J.M. & Zhai, S.. (2026). Normalizing Flows with Iterative Denoising. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:14367-14376 Available from https://proceedings.mlr.press/v306/chen26aj.html.

Related Material