Could you please elaborate more about ICA since in the video “Other techniques” in week 2 [ 1. Machine Learning Modeling Pipelines in Production] at time 5:54, I can’t understand the graph which is used to demonstrate the difference between Ica and pca

Thank you

Hello @Ashish_Sharma6
The graph essentially demonstrates that while PCA (identifies directions in the raw feature space that account for maximum variance.) focuses on finding uncorrelated components, ICA(finds directions that are highly non-Gaussian and statistically independent) goes further to find statistically independent components, which includes removing both correlations and higher-order dependencies.


Great topic. Just found this article explaining the difference:

Happy learning,