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

Thanks for this video It solved my headache

Anyone from 2024 watching this?

Can notes pdf be provided

derivative of log(1-D(G(z))) at D(G(z)) = 0 is -1 so not 0 or dimished, then why the vansihing gradient problem and the need to change the G loss function???

Thank you Ahlad for the beautiful explanation!

This is so good !!!

appreciate your lectures

Thankyou

Thank you Dr Kumar.. Appreciate your efforts very much.... A very useful lecture.. Is there a link to follow along with the pdf notes?

it was very good. thanks

In last part,Summation over p_ij, for all i (error here) it should be for all j.

Ahmad, please let me know how to that in Linux/Ubuntu? You are using an Apple/MacOs machine. Thanks in advance

I’m a bit confused at timestamp 17:16. When we feed the output from the generator, do we freeze the weights of the discriminator, or do we train both networks concurrently?

26:19 Here, I'm not sure if we replace Σ(x) with exp(Σ(x)) really due to it being more numerically stable. I searched Google and found nothing special. The output of the hidden layer in Encoder is (μ, Σ). In the implementation part (the last video in this series), they are "Mean_layer" and "Standard_deviation_layer" variables. The output Σ can be negative because it is the output right off a Fully Connected layer (Dense Layer); however, Standard Deviation of a Distribution can be NEVER negative. To fix this, we simply interpret the "Standard_deviation_layer" (Σ) variable as "log of variance". When needing the variance inside, we simply do "exp(Σ)". I think this is the true motivation of why we need to replace Σ(x) with exp(Σ(x)) at 26:19. Because it is not a "real" variance, but (interpreted as) "log of variance".

Emmanuelle Loop

thanks sir

can you make all video of reinforcement learning free pleaseeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee

this series was a nice review of CNN for me, thanks

Sir I think with the diagram X can not be union of call Yi events. It should be union of some events from Yi.

Valuable lictures. THANK YOU

Magnificent!

Thank you very much. 😊

@19.57 in case 2 you are saying KL divergence (regularizer is present) but in the second figure (for case 2) it is written that 'without regularizer'??/

what is prior distribution? is it the distribution of input data?. If it is, then how can we assume to be gaussian with zero mean and unit variance??

are you missing a summation at 6:02 in the LSTM gradient equation. There are 2 summations in RNN equation but there should be atleast one summation in LSTM equation right? even if we are calculating gradients for one time stamp

Sir can you reply to my mail

thank you so much can you share the code

Thank you so much for this.

Is there a playlist for mathematical preliminary??

sir Kindly finish your NLP playlist.

It looks like Math equations of function code with Python language

Thank you teacher 😊

At 8:35 CS circuit av=-gm1ro2/1+(gm1*(1/gm3//ro3))...Answer At 18:12 Source fillower av=gm1*(RL//(1/gm1)=gm1*RL/(1+gm1*RL).….Answer At 20:59 av=gm1*(1/gm1+1/gmb) approximately gm1*(1/gm1)=1....Answer At 25:26 CG circuit av=vo/vi=-gmvgsRd/-vgs=gm*RD...Answer At 27:05 g_d connect and then i=-gm1vgs=-gm1(0-vin), vin/i= Rin=1/gm1...Answer

How exactly is it Stacked Auto encoder? I could see only one AE was used for one single task which was denoising, Isn't it Denoising AE?

Amazing lecture! Explained everything in very brief, Loved it!

Professor, it will be great if you share the roadmap or the planning of the whole playlist.

Very nicely explained 👍

Professor, later you can start a video series with pytorch. Pytorch playlist is very much needed as less number of people knows it

Awesome lecture

Good to see you back sir 😊😊 after long time We are happy for that ❤

Sir please upload videos fast 🙏

Finally Professor! Thanks for coming back. We need series about LLMs please!

Please share the notebook files also. Your lectures are awesome

is there any other video apart from these 14 video?

sir is this full course?

Reason for filters dimension 9*9

This is not Backprop through time, this is just normal backprop. It does not work on LSTMs or even RNNs as the differential of loss on input weight does not only depend on hidden state, but as each hidden state in time t depends on state in t-1 which in turn depends on input weights again. the differential itself propagates backwards through all time steps. This was the whole point of the paper "Backpropagation through time".

Still best video on VAE after all these years. I rewatch this series whenever i need to brush up VAE