FEM is a python library for solving various combinatorial optimization problems using gradient based mean-field annealing. The current build-in problem types are:

• maximum cut
• balance minimum cut
• maximum k-satisfactory
• vertex cover
• qubo-like instances Some examples can be found on example notebook.

You can also use FEM to solve your own optimization problems as long as the expectation value of the target function can be written as some function of the marginal probability. Please refer to the customize examples for further details.

You can run the jupyter notebooks in benchmarks to reproduce the main results presented in the paper.

One can use conda to install the package with the following commands:

conda env create -f environment.yml

this will create an environment named fem with all the dependencies except for the pytorch, then activate the environment with conda activate fem.

Then pytorch have to be installed manually with

pip3 install torch torchvision torchaudio

see the pytorch website for more details.