Deep Learning 34: (1) Wasserstein Generative Adversarial Network (WGAN): Introduction

[@QueenMate]

This was an amazing mathematical intuition, eagerly waiting for further videos.

[@ahteshamabbasi9503]

One of the best mathematical explanation out there.

[@ankitsingh-xl7bo]

JSD in GAN is only used for the optimal discriminator and not for all the cases... isn't it?

[@MLDawn]

I am a huge fan! First time EVER that I have seen a lecture like this!

[@AhladKumar]

thanks

[@rayll8579]

The video deserves more visibility

[@sangameshkodge1664]

While discussing the KL divergence(5:17) P was unknown and Q was a known distribution. But in the discussion for forward KL(12:16) I see that Q distribution is said to vary (How can Q stretch if Q is already known, shouldn't it be fixed?). Is Q assumed to be unknown and P assumed to be known in the case for forward KL ?

[@pramethgaddale8242]

P is the distribution of the data provided, which is unknown. Q is an approximation to the posterior which is chosen for our convenience. Q is chosen mostly to be gaussian, but if we know the data generating process, we can use different priors too(such as beta or dirichlet, many use Gaussian because it's easy to work with). Coming to Q being to vary, this posterior approximation, forming the ELBO, turns the inference problem to an optimisation. Using SGD, you start with some random mean and variance and work your way to get the best possible ones.

[@darkmythos4457]

Sir, thank you very much for taking the time to make such a greate lectures.

[@marinamaher8211]

Magnificent!

[@vaishanavshukla5199]

wonderful explaination as always

[@AbhishekSen]

Machaya bhai, sab samajh aa gya!!

[@sudarshanregmi14]

Thank you so much. Please, keep making such great videos.

[@iliasaarab7922]

Amazing explanation! Thank you sir! 🙌🏽

[@delseyjohnson3960]

how to get the complete course content

[@asheerali2376]

perfect explaination

[@newtonleibniz879]

Can notes pdf be provided

[@kristianmamforte4129]

amazing lecture!

[@manikantabandla3923]

It could have been more informative if you can point out the relevant papers of lectures.

[@chocclolita]

Thank you so much!

[@appletree6741]

excellent!

[@ayushthada9544]

Sir, this is great video. Not just the explanation was clear but I really like your teaching style too. Could you make a video on StackGANs and Full Bayesian Implementation of GAN too? That would be a big help sir.

[@AhladKumar]

sure

[@ayushthada9544]

@@AhladKumar Thank you so much, sir.

[@ayushthada9544]

@@AhladKumar Thank you so much, sir.

[@spencert94]

If JS Divergence is the only/main reason GANs preform better than VAEs why would you not just use JS Divergence in VAEs. It seems like JS Divergence is just a more stable and symmetrical KL Divergence which doesn't seem like it would actually lead to better results in most cases just be more stable.

[@krishanudasbaksi9530]

maybe coz JS divergence is not differentiable.

[@mostafakattan6971]

This is great job, but can you put subtitle on all your videos.

[@vijetakhare8331]

Superb👌🏼👌🏼

[@satyamdubey4110]

💖💖

[@angquoctien4640]

it would be clearer with subtitle below

[@mimo-wx9mc]

hello sir, I don't understand your English can you add the subtitle that will help me a lot thank you

[@anubhavgupta4917]

Uncut Diamond, Uncut Diamond, Uncut Diamond, Uncut Diamond, Uncut Diamond, Uncut Diamond, Uncut Diamond, Uncut Diamond

[@NancyLee-s5j]

Emmanuelle Loop