search

vitutorial / VITutorial

This repository stores slides for a tutorial on variational inference for NLP audiences.
298 stars 55 forks source link

Normalising flows #33

Closed philschulz closed 5 years ago

philschulz commented 5 years ago

This is still work in progress, so no need to merge yet. I've cleaned up the notation for density estimation. I need to work further on inference. The overall structure is complete, though. I will add more pictures to the intro. Let me know what you think.

Important: check the vi_marcros.sty for the \abs and \jacob macros. Let's use these everywhere and then we can adjust them as needed.

philschulz commented 5 years ago

It's complete now. I still need to change the macros but the content is there.

philschulz commented 5 years ago

I prettied it up and am using your macros. The picture in the intro are missing but the use cases are in good shape. Indexing the variables is a bit tricky. The convention is: h_i maps input i-1 to i and is computed by function g^{(i)}. The sequence is from \epsilon to the target variable. That means x = \epsilon^{(K)} and z^{(1)} = \epsilon^{(1)}.

philschulz commented 5 years ago

Done! Complete with pictures.

philschulz commented 5 years ago

I've fixed the inverses and the joint model. Can you review and merge, please?

philschulz commented 5 years ago

Thanks for the feedback. I'll update later today.

philschulz commented 5 years ago

Added an illustration for NF sampling and rebased on master. Also addressed all your other comments.

wilkeraziz commented 5 years ago

Thanks for lecture and the edits! I look forward to using it here at UvA as well :D