Decohering tensor network QML models

Unitary Tree Tensor Network and MERA

Efficient unitary tree tensor network (TTN) and multi-scale entanglement renormalization ansatz (MERA) built with TensorFlow, with tunable local dephasing channels at every layer of the tensor networks and tunable number of ancillas, benchmarked on compressed MNIST, KMNIST, and Fashion-MNIST. Code comments included.

To setup,

git clone https://github.com/HaoranLiao/dephased_ttn_mera.git
cd dephased_ttn_mera/
conda create -n tnqml python=3.8
conda activate tnqml
pip install -r requirements.txt

The compatible tensorflow versions should be around 2.4 - 2.7. For Apple M1, do the following,

conda install -c conda-forge tensorflow==2.6.0
conda install -c conda-forge tensorflow==2.4.0

Add this to ~/.bashrc replacing the HOME_DIR:

export PYTHONPATH="${PYTHONPATH}:HOME_DIR/dephased_ttn_mera/"

and source ~/.bashrc.

To use GPUs,

pip install -r requirements-gpu.txt

$~$

To run the MERA, cd dephased_ttn_mera/mera.

Configure config_example.yaml, and run python model.py

To run the unitary TTN, cd dephased_ttn_mera/uni_ttn/tf2.

Configure config_example.yaml, and run python model.py

$~$

The main scripts to construct the tensor networks and to do the training are:

• data.py (for dataset loading and preprocessing),
• model.py (define the workflow),
• network.py (construct the network),
• under the folders cd dephased_ttn_mera/uni_ttn/tf2 for unitary TTN and cd dephased_ttn_mera/mera for MERA

$~$

Using the code please consider citing:

 @article{Liao_2023,
  title={Decohering tensor network quantum machine learning models},
  author={Liao, Haoran and Convy, Ian and Yang, Zhibo and Whaley, K. Birgitta},
  journal={Quantum Machine Intelligence},
  volume={5},
  number={1},
  pages={7},
  year={2023},
  publisher={Springer},
  doi={https://doi.org/10.1007/s42484-022-00095-9},
}

Reference: Liao et al., Decohering Tensor Network Quantum Machine Learning Models, Quantum Mach. Intell. 5(1), 7 (2023)