Ali Ghodsi, Lec 12: Neural Networks, Autoencoders, Word2Vec

[@EvanZamir]

Great lecture. Props to your cameraman for following you.

[@aze90]

Great teacher!

[@kathirs1]

Awesome explanation sir.. Make more videos please

[@gautamkarmakar3443]

Awesome lectures!! Is there anyhow I can get the slides?

[@rauldearaujo7188]

Great explanation! Thank you! :D

[@adels1388]

I think Vw is a column of W' not a row of it. am I right?

[@mubintirsaiwala7141]

Yes, you're right.

[@gregarityNow]

seconded

[@EzraSchroeder]

@26:16 why can we not guarantee that the output vector is sparse at test time?

[@leeris19]

Actually, I think we can guarantee it if we train the network enough but that is actually a bad outcome as it suggests overfitting. Ideally, we want the outcome to be not that sparse in order for the model to generalize better. The model should have this notion that "I know this is what should happen, but it's still possible that other results may happen".

[@gregarityNow]

at 44:57 you clandestinely added a minus. I think that should not be there if you are maximizing!

[@crazybrembo]

Is Ui a similarity at i-th word?