This is a course project for ECE594BB (Selected Topics in High-Dimensional Tensor Data Analysis) instructed by Dr. Zheng Zhang at University of California, Santa Barbara. This project aims to utilize the Google TensorNetwork toolbox to perform machine learning over the tree tensor network inspired by quantum computing. This project is mainly based on the following work:
[1] Ding Liu, Shi-Ju Ran, Peter Wittek, Cheng Peng, Raul Blázquez García, Gang Su, and Maciej Lewenstein. Machine learning by unitary tensor network of hierarchical tree structure, New Journal of Physics, 21(7), 073059, 2019.
Their code is also available on Github link.
The code is Python3-based, and the following packages are required to run the repository.
- Download the entire repository
- Copy corresponding data files from this link, and the put the content into a folder called
data.
To reproduce the results from [1], the following command can be executed.
./qmle reproduce --data-path ./data
To run the experiments using TensorNetwork, the following command can be executed.
./qmle run --data-path ./data --dataset MNIST
Please check the arguements of the program with
./qmle reproduce -h
or
./qmle run -h
There are mainly a few groups of arguments.
- Arguments for logging and output directory:
prefix,log,log-level - Arguments for input:
data-path,dataset - Arguments for paper reproducing hyperparameter:
num-epoch,bond-data,bond-inner,num-train-single,num-test-each
For more details, please check utils/arg_parse.py file.
- Test different hyperparameter setting for the paper reproducing, and collect the results (Jose Acuna)
- Try more sophiticated normalization method for the image, please check function
image_normalizationin./third-party/ttn_ref.py. (Jose Acuna) - Use TensorNetwork to implement the entire flow, and generalize for more datasets (Hanbin Hu)