Code for paper Low-rank and Orthogonal Space for Weights Modification Enhances Learning Efficiency in Neural Networks without Gradient Backpropagation.
If the environment configurations are different, the results may vary greatly or even failing to work properly. You may need to adjust the code details according to your needs. Linux: Ubuntu 16.04
-
cuda 11.8
-
Python 3.7.16
-
torch 1.13.1 (pytorch)
-
torchvision 0.14.1
-
numpy 1.21.4
-
scikit-learn 1.0.2
-
scipy 1.7.3