CS480/680 Lecture 18: Recurrent and recursive neural networks

[@VahidOnTheMove]

It's unbelievably easy to understand the lecture. The best explanation I have seen.

[@tahamagdy4932]

Teachers should strive to be like you. Thank You

[@gregpaokfc]

Great lecture,really insightful and more relevant than ever. Thank you mr. Poupart.

[@notanape5415]

Really amazing explanation for Attention! Thanks for making this lecture public.

[@datafiasco4126]

Great job in explaining the concepts. Really liked the roll out diagrams.

[@hadisjalesiyan1558]

Finally, I understood Attention. Thank you so much

[@vitocorleone1991]

Awesome, God bless you, professor.

[@user-or7ji5hv8y]

Best explanation on attention so far for understanding the intuition.

[@rubenpartono]

10:21 Being unfamiliar with how automatic differentiation works, this feels absolutely magical.

[@micknamens8659]

1:33:40 A mathematical graph has no fixed geometrical layout. Hence the distriction of functions u and v for "left" and "right" child seems odd.

[@datafiasco4126]

This is more akin to a tree in computing and hence legacy terminology is used as left and right child.

[@andrewmeowmeow]

Great lecture!

[@user-or7ji5hv8y]

Does s3 depending on s2 and the convex combination of all of the h create unnecessary duplication, since s2 also depend on all the of h?

[@samlaf92]

50:00 Is this really an LSTM? Where's the "c" cell? Seems like a weird mix of LSTM and GRU...

[@samlaf92]

Nevermind. Should have waited until 59:00. I guess the "simple" version presented first around 50:00 would best be described as a peephole LSTM?

[@user-or7ji5hv8y]

I don't see how under HMM, why the Y depend on X, given the direction of the arrow. Does the states depend on the past observations under HMM?

[@thak456]

yes

[@puyncharasr6963]

Thumb up