PCA Based Model Modification

Dependencies

• python == 3.7
• pytorch == 1.12.1
• numpy == 1.21.5
• thop == 0.1.1

Implementation

Compute_PCA and Extract Primary Layers

• Calculate PCA explained variance ratio from compressed output for each activation layer
• Please check out section 1 and 2 in quantization_conversion.ipynb

Training Script

• Currently available for Cifar10/Cifar100 training for vgg11/13/16/19 Details in train.py

        python train.py --pretrained --dataset cifar10 --model vgg13

quantization

• Please check out quantization_conversion.ipynb Detailed instructions are to be added