Training of Quantum Boltzmann Machines

Library for training QBMs based on quimb, a popular python library for quantum information and many-body physics, with MPI acceleration and Tensor Network methods.

Setup

The package environment is handled by poetry which will install all the dependencies and the package when running poetry install in the root folder of the project.

Benchmark

The benchmarking script can be run with poetry run python scripts/benchmarking.py. A list of command line arguments is shown below:

usage: benchmark.py [-h] [--n N] [--t T] [--b B] [--l L] [--dn DN] [--lr LR] [--e E] [--er ER] [--sn SN] [--qre] [--pre_l PRE_L] [--pre_lr PRE_LR] [--pre_e PRE_E] [--seed SEED] [--output OUTPUT]

Train a QBM model to represent a target Gibbs state

options:
  -h, --help       show this help message and exit
  --n N            Number of qubits (4)
  --t T            Label of target model (0)
  --b B            Inverse temperature of target model (1.0)
  --l L            Label of QBM model (0)
  --dn DN          Intensity of depolarizing noise (0.0)
  --lr LR          Learning rate (None)
  --e E            Number of traninig epochs (1000)
  --er ER          Error tolerance for gradients (1e-6)
  --sn SN          Standard deviation of gaussian shot noise for computing gradients (0.0)
  --qre            If we want to compute and output relative entropies
  --pre_l PRE_L    Label of QBM model for pretraining (None)
  --pre_lr PRE_LR  Learning rate for pretraining (None)
  --pre_e PRE_E    Number of traninig epochs for pretraining (300)
  --seed SEED      Seed for PRNG (1)
  --output OUTPUT  Output for data and figures (data/)

Data

The data folder includes already some results from training different QBMs on Gibbs states for 5 different Hamiltonians:

• 1D Heisenberg model
• 1D Hubbard model
• 1D Transverse Field Ising model
• 2D Hubbard model
• J1-J2 spin glass model

More information about the Hamiltonians can be found in the file hamiltonians.py.

Cite

This package is jointly developed by Panasonic and Quantinuum and distributed under Apache-2.0 license. If you use this code in your research, please cite it using the following:

@misc{qbm-benchmark-dataset-2024,
  author = {Enrico Rinaldi, Yuta Kikuchi, Ryuji Sakata},
  title = {Quantum Boltzmann Machine training and benchmarking dataset},
  year = {2024},
  note = {GitHub repository},
  howpublished = {\url{https://github.com/CQCL/qbm_benchmark_dataset}},
}