We have modified the code of the Github repository: https://github.com/allenai/hidden-networks.git 
Its original license is attached. 

Our modifications include: 
- an extension of the He parameter initialization scheme by a non-zero bias initialization (which is set as default in the current implementation);
- pruning of biases together with the network weights;
- two added model architectures (univ.py), for which we have proven that they contain universal lottery tickets; 

The two MNIST experiments that we have described in the main manuscript can be startet by running

Polynomials:
python main.py --config path/to/folder/edge-popup/configs/poly.yml --prune-rate 0.5 --initBias "kn-nonzero-bias"

Fourier:
python main.py --config path/to/folder/edge-popup/configs/fourier.yml --prune-rate 0.5 --initBias "kn-nonzero-bias"
