Probabilistic Tensor Train Completion based on Gaussian-product-Gamma Prior, with Automatic Rank Estimation
- demo_rank_accuracy_test.m
Run this demo to test the accuracy of the rank estimation. It randomly generate noise 10 times and tests VITTC for different settings.
- demo_image_completion.m
Run this demo to test VITTC on image completion. The provided image in "TestImages/missing_rate80_1.mat" is the 'jellybeans' image.
- f_tensorfolding/
Includes functions for ket-folding and the improved ket-folding in the reference.
- f_vittc/
Includes functions realizing the variational inference algorithms proposed for the probabilistic tensor train model.
- f_datageneration/
Includes functions that generate synthetic data.
- f_perfevaluate/
Includes functions that evaluate the performance of the recovered tensor.
- TestImages/
A 'jellybeans' image, and a mask with 80% entries missing
- experiment results/
A folder used for storing results.
[1] Xu, L., Cheng, L., Wong, N., & Wu, Y. C. (2020). Learning tensor train representation with automatic rank determination from incomplete noisy data. arXiv preprint arXiv:2010.06564.
[2] Xu, L., Cheng, L., Wong, N., & Wu, Y. C. (2021, December). Overfitting Avoidance in Tensor Train Factorization and Completion: Prior Analysis and Inference. In 2021 IEEE International Conference on Data Mining (ICDM) (pp. 1439-1444). IEEE.
[3] Xu, L., Cheng, L., Wong, N., & Wu, Y. C. (2023). Tensor train factorization under noisy and incomplete data with automatic rank estimation. Pattern Recognition, 141, 109650.