pytorch
need to implement the CKA function for model interpretability and understanding
🚀 Feature
The Centered Kernel Alignment (CKA) is function to compare two neural networks or the layers, it can be used to explain model dynamics, and it is widely used for understanding the representations learned by neural networks.
Motivation
see: https://github.com/AntixK/PyTorch-Model-Compare
Pitch
hope the CKA is implement in the captum project.
@joshua-xia, we have an implementation of CKA but we haven't open sourced it because of other high priority projects. It might take some time before we open source it.
@joshua-xia, we have an implementation of CKA but we haven't open sourced it because of other high priority projects. It might take some time before we open source it.
Thanks a lot! Would you please open source it for me, we currently using CKA in our project, so that I think we can test the implementation for the community.