This readme file is an outcome of the CENG501 (Spring 2021) project for reproducing a paper without an implementation. See CENG501 (Spring 2021) Project List for a complete list of all paper reproduction projects.
In this project we tried to implement the method proposed in the Channel Pruning Guided by Classification Loss and Feature Importance paper from Association for the Advancement of Artificial Intelligence,2020. In this paper, writers proposed new method for pruning channels of a CNN network taking classification loss and feature importance into account. Different from methods based on layer-by-layer prunning, writers add the classification loss for a certain channel before deciding tobe pruned or not.
This paper proposes new method for channel prunning based on feature importance and classification loss. Different from other layer-by-layer methods writers added the contribution of channel to the loss and feature importance term to the objective functio which is an optimization problem solved in 2 steps.
2.1. Unfortunately we could not complete the method for the deadline but we will work on it later on. Only finetuning VGG13 is implemented for now.
Explain the original method.
Explain the parts that were not clearly explained in the original paper and how you interpreted them.
Describe the setup of the original paper and whether you changed any settings.
Explain your code & directory structure and how other people can run it.
Present your results and compare them to the original paper. Please number your figures & tables as if this is a paper.
Discuss the paper in relation to the results in the paper and your results.
[1]Jinyang Guo, Wanli Ouyang, Dong Xu; Channel Pruning Guided by Classification Loss and Feature Importance, Association for the Advancement of Artificial Intelligence, 2020. https://arxiv.org/pdf/2003.06757.pdf
Provide your names & email addresses and any other info with which people can contact you.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent