Deep Learning 29: (3) Generative Adversarial Network (GAN) : Explanation of Loss Function

In this lecture we will gain more insights into the Loss function of Generative Adversarial Networks
#adversarial#generative#deeplearning

[@mrtandon5278]

I invested 40k in Udacity Deep Learning Nanodegree program, followed by 32k investment in applied ai course. But nobody bothers themselves to teach GAN mathematics concepts. On the other side, a great teacher digs deep in the ocean just to make their student understand what actually GAN is. You left no stone unturned. Thanks for sharing your knowledge and for your efforts.

[@wolfisraging]

Literally the best explanation of gans on whole KZbin

[@prateekthakur4060]

I am with you man. One of the best explanation on youtube.

[@nisanchhetri1162]

@@prateekthakur4060 do watch this then you will retake this statement. kzbin.info/www/bejne/eoKxf4CfdrVootk

[@nisanchhetri1162]

do watch this then you will retake this statement. kzbin.info/www/bejne/eoKxf4CfdrVootk

[@anshumishra4193]

@@nisanchhetri1162, it is in Chinese. The content looks excellent, but it's tough to follow just looking at the subtitles.

[@superaluis]

Only lecture on KZbin with the advanced concepts all throughly explained to the detail. Subscribed

[@rachitverma9088]

Your attention to detail regarding the concepts is truly remarkable. Thanks a lot for making this excellent series!

[@user-ky7bi9cl2z]

Only KZbin channel which gives proper insights to Deeplearning.

[@MLDawn]

are you for real?! Keep it up please! It is like you know where I had problem understanding it, and you dissect that very part beautifully!

[@shineshine9599]

True...the best explanation...thank u so much!!

[@NeerajAithani]

It's been 10 months I implemented GAN and it was easy. I tried to read paper didn't understand. finally, your lectures helped me to understand GANs fully, probably the best explanation on whole internet :).

[@Gauravkr0071]

neeraj how do u implement it, i am searching for how to implement it on keras or any other librray

[@NeerajAithani]

@@Gauravkr0071Dewangan there are plenty of resources on the internet to implement it. If you are a beginner I would suggest you first get comfortable with the library you are using and build some small models. After you feel comfortable then go to advance topics like GAN.

[@Gauravkr0071]

@@NeerajAithani ok sir , just one thing , there is so much maths involved in it , but when i see the code of any GAN , there is no such maths involved in it, in the code

[@hiteshnagothu887]

I really appreciate all the series explained so well without giving up on the foundation of Maths, of course, it's all the game of Maths! Looking forward for more such content.

[@puspaksahu3342]

Love from IISc pure maths ! salute , no one cares to go into mathematical detail as yours on utube

[@elyseemanimpiregasana2117]

I don't know what I can say. But always in deep learning, If I have a presentation related to the topics he explained, I have to take the course. After that, everyone takes me as a specialist in the topic. you really deserve respect.

[@bosepukur]

fantastic lecture series ....I am hooked

[@shail12aksh]

Thoroughly explained the whole concept of GAN...Thanks for explaining GAN in such an understandable manner...

[@rahulrai1329]

Thanks a lot. Very nice and intuitive explanation of GAN and very useful.I request you to Please upload more and more content regarding GAN and latest papers.

[@akshitasood6455]

This explanation makes the concept of GANs so much easier !!! Great Lecture !!!

[@vivek65522]

Thank you so much Prof. You have done a great job explaining. Beauty of the concepts lie in the details of mathematics which you have to explained to the core.

[@jitendradhiman5815]

The intuition behind GAN is very nicely explained. Thank you sir for bringing such informative videos to us.

[@advaittilak614]

I wish I could personally meet and thank you Alhad Sir.

[@doublesami]

I'm very motivated sir after reading about you. Following your lectures from a long time . Your methodology is best one . Best wishes Just keep motivated us by providing quality literature reviews

[@fawadkhan2955]

This is great. Liked and Subscribed. Please keep making more videos like this on the most common deep-learning techniques. Suggestion: CycleGAN

[@aashwinsharma1859]

The efforts just to explain small concepts and proofs.... 🙏

[@RAJANKUMAR-mi1ib]

wow....you made me understand each and every concept....keep the good work going!

[@ssshukla26]

Nobody explained this ... Not even my college professor ... Thank you sir ..

[@ahmedheakl5181]

You are the best youtuber ever

[@soumambanerjee1816]

You have the power to generate Netflix alike addictive content in deep learning ❤️

[@baabakasadi5440]

Beautiful explanation

[@msic1376]

Sir, this is the best explanation of GANs on youtube by far! Only one doubt: why at 12:05 we take the derivative only of the argument of the integral and not of the whole integral?

[@ashj1165]

Started loving maths again with these explanations!

[@gauravsahani2499]

Best Explaining Sir, Thankyou so much for this Playlist!

[@kwippo]

3:17 In equation 2, is it supposed to be D(G(z)) instead of D(G(x))?

[@sugamtyagi101]

Thank you so much for going into the details.

[@solitaryreaper1363]

That's a really good explanation sir, thank you so much.

[@MuhammadSaleem-hc6ng]

Good Insight

[@mariasargsyan5170]

Hi, thanks for explaining the change of variables in here. However, it is generally not true that the G is an invertible function, especially in the context of Deep Neural Networks, right or am I missing something?

[@CarlosGarcia-hs8yg]

best video ever. Thanks a lot man!

[@karimmache4018]

Very interesting lecture, but please make your notation clear.

[@aakanksha7877]

thanks sir, you made it relatiively simple and effective.

[@prateekthakur2519]

After reading so many blogs, papers, videos and paying for courses... finally I found the best explanation!!! Thank you so much! Just one question while take derivative of integrad why did we not consider the derivative of PDFs Pdata(x) and Pg(x) ?

[@elyseemanimpiregasana2117]

You know to change the output we use gradient descent. In that case, we differentiate the loss with respect to the parameters of the Generator and Discriminator. No need for a probability derivative, I hope it is understandable, for any misunderstanding let me know.

[@nadarasarbahavan6269]

Amazing explanation. I am telling everybody

[@bSharpHacker]

What happened to the x's inside the integrals of equation 1, after combining 1 & 2?

[@amarnathjagatap2339]

Best explanation sir.. Thank you sir

[@ayanmaiti1789]

i did not get how to convert instance to sample? that is how we arrived at the expectation terms ? any pointers to good reference?

[@superaluis]

I too got a little bit confused on this part... If you find on internet any explanation about this part please share here.

[@pareshupadhyay297]

Thank you for the amazing explanation .I have one query that is-- You said that random variable x has a distribution Pdata(x) and at time 9:25 you said that x=G(z), but this only be true after many iteration, my question is why this assumption is true at the starting.

[@elyseemanimpiregasana2117]

I think you fix it as a target to achieve since our wish is to generate output similar to the existing data that will fool the discriminator.

[@ankurgupta2806]

what can be the intuitive explanation of the change between the two integrals?

[@pratik9948]

can u make a video on gan loss functions that is which loss function to be used for generator and disriminatoor and also combine model which to use and why it is hard to train gan thanks in advance

[@ivanrezic]

Hello, great video! Can you please point me to the proof of that transformation formula used at 7:29?

[@AhladKumar]

see Dependent variables and change of variables at the following link en.wikipedia.org/wiki/Probability_density_function

[@bosepukur]

In equation 2 should it not be D(G(x)) ----> D (G(z))

[@AhladKumar]

right..thanks for pointing out

[@Gauravkr0071]

yes

[@taureanamir]

the explanations are amazing. thanks tonnes.

[@prarabdh6295]

thanks