The core function modules, including:
Code implementation of Algorithm 1 in Appendix B.1.
Code implementation of Algorithm 2 in Appendix B.1.
Code implementation of Algorithm 3 in Appendix B.2.
Code implementation of Algorithm 4 in Appendix B.3 and Fully-Connected Network.
Code implementation of Standardization Network, a part of Algorithm 4. See formula (12)-->(13).
Code implementation of VC dimension based loss, a part of Algorithm 4. See formula (14).
Code implementation of some data processing modules, including: load data from file and generate epoch, batch, etc.
including:
An experiment on Taichi data set, 628+629=1257 training samples, Gaussain kernel, 2000 kernel parameters.
An experiment on MNIST data set, 10x10=100 training samples, 10000 test sample, FCNet.
When VC_Dim_Loss_weight>0, the proposed boosting module will work.
When VC_Dim_Loss_weight=0, the proposed boosting module will not work.
Two data sets, including:
Files of TaiChi data set.
Files of MNIST data set.
The default root folder of result files, including:
Result files of experiments on Taichi data set.
Result files of experiments on MNIST data set.
This file.
Ubuntu 18.04.5
PyCharm 2020.1.5
Python 3.8.12
PyTorch 1.7.0
Numpy 1.21.2
hdf5storage 0.1.16
scipy 1.7.1
etc.