[Blog] Notes on Weight Initialization for Deep Neural Networks

Hi everyone,

I’ve expanded on the contents of the weight init notebook at https://madaan.github.io/init/

I’ve added proofs for some of the results mentioned in the course notebook, and have tried to stress on some aspects of the discussion.

The discussion at the relevant pull request may also be interesting.

I have not linked to the specific notebook in the blog, and have only acknowledged fast ai in general because the course is not public yet. I’ll make the change after the notebooks are released to the public.

Please feel free to suggest improvements/point out issues.