Deep Learning(CS7015): Lec 9.2 Unsupervised pre-training

[@raghaviyer3097]

This lecture is SUPER under rated!

[@ashhadtoseef1834]

this is the best explanation out there is.

[@fxexile]

This is just mind blowing

[@silentgoogler9471]

Fantastic lecture. Thanks for sharing!

[@linahuang94]

This is great, thank you!

[@swaroopmeher2995]

Thank you!

[@anuragdathatreya1598]

Can you also tell us why is there a square for L1 and a circle for L2?

[@rajatkumarsingh8523]

It's because for L1, we have omega(theta) as |theta|. So if you think optimization problem as minimizing Mean squared error subject to some constraint |theta| < s. Then the graph for |theta| < s will be the diamond region. Similarly for L2, constraint will be (|theta square) < s, which is region inside circle given by equation (theta square) = s (that means centre of circle is origin)

[@umang9997]

@@rajatkumarsingh8523 what is 's'?

[@aniketsukhija9916]

can someone answer the questions at 10:58