DPaI: Differentiable Pruning at Initialization with Node-Path Balance Principle

Requirements

Tested on python==3.10.14, pytorch==2.4.0, cuda==11.8, torchvison==0.19.0, numpy==1.26.4

alpha: Trade-off between the number of effective nodes/kernels and effective paths. A higher alpha results in a greater number of effective nodes/kernels (0.0 <= alpha <=1.0).
beta: Trade-off between the number of effective kernels and effective nodes. A higher beta results in a greater number of effective kernels (0.0 <= beta <=1.0).
lr_score: The learning rate for the score parameters.
num_steps: The number of steps for updating the score parameters

bash experiments/run.sh

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
experiments		experiments
.gitignore		.gitignore
README.md		README.md
efficientnet.py		efficientnet.py
main.py		main.py
metric.py		metric.py
models.py		models.py
npb.py		npb.py
scheduler.py		scheduler.py
utils.py		utils.py