Multiple Texture Boltzmann Machines

Jyri Kivinen, Christopher Williams
; Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics, PMLR 22:638-646, 2012.

Abstract

We assess the generative power of the mPoT-model of [10] with tiled-convolutional weight sharing as a model for visual textures by specifically training on this task, evaluating model performance on texture synthesis and inpainting tasks using quantitative metrics. We also analyze the relative importance of the mean and covariance parts of the mPoT model by comparing its performance to those of its subcomponents, tiled-convolutional versions of the PoT/FoE and Gaussian-Bernoulli restricted Boltzmann machine (GB-RBM). Our results suggest that while state-of-the-art or better performance can be achieved using the mPoT, similar performance can be achieved with the mean-only model. We then develop a model for multiple textures based on the GB-RBM, using a shared set of weights but texture-specific hidden unit biases. We show comparable performance of the multiple texture model to individually trained texture models.

Cite this Paper


BibTeX
@InProceedings{pmlr-v22-kivinen12, title = {Multiple Texture Boltzmann Machines}, author = {Jyri Kivinen and Christopher Williams}, booktitle = {Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics}, pages = {638--646}, year = {2012}, editor = {Neil D. Lawrence and Mark Girolami}, volume = {22}, series = {Proceedings of Machine Learning Research}, address = {La Palma, Canary Islands}, month = {21--23 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v22/kivinen12/kivinen12.pdf}, url = {http://proceedings.mlr.press/v22/kivinen12.html}, abstract = {We assess the generative power of the mPoT-model of [10] with tiled-convolutional weight sharing as a model for visual textures by specifically training on this task, evaluating model performance on texture synthesis and inpainting tasks using quantitative metrics. We also analyze the relative importance of the mean and covariance parts of the mPoT model by comparing its performance to those of its subcomponents, tiled-convolutional versions of the PoT/FoE and Gaussian-Bernoulli restricted Boltzmann machine (GB-RBM). Our results suggest that while state-of-the-art or better performance can be achieved using the mPoT, similar performance can be achieved with the mean-only model. We then develop a model for multiple textures based on the GB-RBM, using a shared set of weights but texture-specific hidden unit biases. We show comparable performance of the multiple texture model to individually trained texture models.} }
Endnote
%0 Conference Paper %T Multiple Texture Boltzmann Machines %A Jyri Kivinen %A Christopher Williams %B Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2012 %E Neil D. Lawrence %E Mark Girolami %F pmlr-v22-kivinen12 %I PMLR %J Proceedings of Machine Learning Research %P 638--646 %U http://proceedings.mlr.press %V 22 %W PMLR %X We assess the generative power of the mPoT-model of [10] with tiled-convolutional weight sharing as a model for visual textures by specifically training on this task, evaluating model performance on texture synthesis and inpainting tasks using quantitative metrics. We also analyze the relative importance of the mean and covariance parts of the mPoT model by comparing its performance to those of its subcomponents, tiled-convolutional versions of the PoT/FoE and Gaussian-Bernoulli restricted Boltzmann machine (GB-RBM). Our results suggest that while state-of-the-art or better performance can be achieved using the mPoT, similar performance can be achieved with the mean-only model. We then develop a model for multiple textures based on the GB-RBM, using a shared set of weights but texture-specific hidden unit biases. We show comparable performance of the multiple texture model to individually trained texture models.
RIS
TY - CPAPER TI - Multiple Texture Boltzmann Machines AU - Jyri Kivinen AU - Christopher Williams BT - Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics PY - 2012/03/21 DA - 2012/03/21 ED - Neil D. Lawrence ED - Mark Girolami ID - pmlr-v22-kivinen12 PB - PMLR SP - 638 DP - PMLR EP - 646 L1 - http://proceedings.mlr.press/v22/kivinen12/kivinen12.pdf UR - http://proceedings.mlr.press/v22/kivinen12.html AB - We assess the generative power of the mPoT-model of [10] with tiled-convolutional weight sharing as a model for visual textures by specifically training on this task, evaluating model performance on texture synthesis and inpainting tasks using quantitative metrics. We also analyze the relative importance of the mean and covariance parts of the mPoT model by comparing its performance to those of its subcomponents, tiled-convolutional versions of the PoT/FoE and Gaussian-Bernoulli restricted Boltzmann machine (GB-RBM). Our results suggest that while state-of-the-art or better performance can be achieved using the mPoT, similar performance can be achieved with the mean-only model. We then develop a model for multiple textures based on the GB-RBM, using a shared set of weights but texture-specific hidden unit biases. We show comparable performance of the multiple texture model to individually trained texture models. ER -
APA
Kivinen, J. & Williams, C.. (2012). Multiple Texture Boltzmann Machines. Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics, in PMLR 22:638-646

Related Material