Hacker News new | ask | show | jobs
by vanilla-latte 1126 days ago
For me recently, it was learning about backpropagation with batch normalisation.

I realised that I had many gaps understanding what and how to partially derive with vectors and making sure their dimensions aligned with the 'with respect to variable'.

Most of my exposure was just to 'variables/letters' and not vectors. So thinking in dimensions during derivation caught me off guard.

Edit: It took my 3 days on and off to figure it out...