Code for SpecNet2
Folder 'batch_run' stores the data and MATLAB code for figure generations; folder 'python' contains Python files for SpecNet2 (and SpecNet1); folder 'matlab' contains MATLAB files for the linear algebra version of SpecNet2 (and SpecNet1).
Packages for the python virtual environment are listed in "./python/requirement.txt". To train a network, run the command -- python main.py para.json (or mnist_para.json). Network architecture and hyperparameters are encoded in the json files.
Reference: Z. Chen, Y. Li, and X. Cheng. "SpecNet2: Orthogonalization-free spectral embedding by neural networks." The Third Mathematical and Scientific Machine Learning Conference (MSML 2022).
Contact ziyuchen@umass.edu if you have any issue to run the code.