Kepler codebook

Junrong Lian, Ziyue Dong, Pengxu Wei, Wei Ke, Chang Liu, Qixiang Ye, Xiangyang Ji, Liang Lin
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:29511-29530, 2024.

Abstract

A codebook designed for learning discrete distributions in latent space has demonstrated state-of-the-art results on generation tasks. This inspires us to explore what distribution of codebook is better. Following the spirit of Kepler’s Conjecture, we cast the codebook training as solving the sphere packing problem and derive a Kepler codebook with a compact and structured distribution to obtain a codebook for image representations. Furthermore, we implement the Kepler codebook training by simply employing this derived distribution as regularization and using the codebook partition method. We conduct extensive experiments to evaluate our trained codebook for image reconstruction and generation on natural and human face datasets, respectively, achieving significant performance improvement. Besides, our Kepler codebook has demonstrated superior performance when evaluated across datasets and even for reconstructing images with different resolutions. Our trained models and source codes will be publicly released.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-lian24a, title = {Kepler codebook}, author = {Lian, Junrong and Dong, Ziyue and Wei, Pengxu and Ke, Wei and Liu, Chang and Ye, Qixiang and Ji, Xiangyang and Lin, Liang}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {29511--29530}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/lian24a/lian24a.pdf}, url = {https://proceedings.mlr.press/v235/lian24a.html}, abstract = {A codebook designed for learning discrete distributions in latent space has demonstrated state-of-the-art results on generation tasks. This inspires us to explore what distribution of codebook is better. Following the spirit of Kepler’s Conjecture, we cast the codebook training as solving the sphere packing problem and derive a Kepler codebook with a compact and structured distribution to obtain a codebook for image representations. Furthermore, we implement the Kepler codebook training by simply employing this derived distribution as regularization and using the codebook partition method. We conduct extensive experiments to evaluate our trained codebook for image reconstruction and generation on natural and human face datasets, respectively, achieving significant performance improvement. Besides, our Kepler codebook has demonstrated superior performance when evaluated across datasets and even for reconstructing images with different resolutions. Our trained models and source codes will be publicly released.} }
Endnote
%0 Conference Paper %T Kepler codebook %A Junrong Lian %A Ziyue Dong %A Pengxu Wei %A Wei Ke %A Chang Liu %A Qixiang Ye %A Xiangyang Ji %A Liang Lin %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-lian24a %I PMLR %P 29511--29530 %U https://proceedings.mlr.press/v235/lian24a.html %V 235 %X A codebook designed for learning discrete distributions in latent space has demonstrated state-of-the-art results on generation tasks. This inspires us to explore what distribution of codebook is better. Following the spirit of Kepler’s Conjecture, we cast the codebook training as solving the sphere packing problem and derive a Kepler codebook with a compact and structured distribution to obtain a codebook for image representations. Furthermore, we implement the Kepler codebook training by simply employing this derived distribution as regularization and using the codebook partition method. We conduct extensive experiments to evaluate our trained codebook for image reconstruction and generation on natural and human face datasets, respectively, achieving significant performance improvement. Besides, our Kepler codebook has demonstrated superior performance when evaluated across datasets and even for reconstructing images with different resolutions. Our trained models and source codes will be publicly released.
APA
Lian, J., Dong, Z., Wei, P., Ke, W., Liu, C., Ye, Q., Ji, X. & Lin, L.. (2024). Kepler codebook. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:29511-29530 Available from https://proceedings.mlr.press/v235/lian24a.html.

Related Material