Видео

Question on slide 12: As I understand it, we do know what the denoising process would be for an image. Just substract the noise that was added to generate the noisy version. But we want a NN to learn these parameters so that once it's trained, we can feed it random noise and it will output novel images. Is this statement correct?

thank you so much🤎🤎🤎

Great video!

@34:34 - The last integral on the RHS, why is there an S_theta(x) "square" term ? .. shudn't it just be s_theta(x) ??

It should be -y.x if y.f(x) < 1

On page 16 (around 38minutes): "The probability is non convex". I think the important thing is that the problem is non convex on Mu and sigma. It is true that P(x) is non-convex but this wouldn't be a problem if it was convex on mu/sigma while non convex on X. I got a little confused by the image that shows the mixture is non convex on the x axis and wanted to clarify if anyone also finds useful :) (and hope it makes sense)

the f-divergence part of the lecture is NOT very clear ... :(

great lecture but i wonder the slides look an awful lot like the standford cs236 lecture slides did you develope the course together 🙈

What is meant by parameters? How do you calculate the number of parameters? After 25:56 it is mentioned number of parameters is 2^n-1

Well explained

H(1-\epsilon)^n isn't a probability as is unbounded. To make it a probability must be defined as min(1, H(1-\epsilon)^n).

Great lecture!

I havent seen the best course like that taking from basics, I love this !!

Where to find the assignments? They are absent in the website.

Hi professor, once Neural Autoregressive model(NADE) is trained or for that matter any model like wavenet or pixel CNN, How do i compute the probability of a new image belonging to learnt distribution?

Definitely one of the best courses on the topic !

Can you recommend any specific books that we can follow alongside this course ?

Excellent ! By far the Bestest lecture series on Generative Models simply because it explains the intuition behind all the underlying math .. just fabulous !!

Awesome !

Loving it !! ... Can't thank enough ..

Terrific Lectures ... just bang on the target .. addresses all the issues which are a bit tricky to understand .. Amazin !.. Thanks a ton

Excellent ! .. Thanks ! .. addresses almost all of the pain points pertaining to Probabilitistic Machine Learning

Turing out to be just amazing set of lectures .. exactly touching upon the most painful (to understand) points in "probabilistic machine learning" .. awesome !

some feedback : very shallow , non concrete throw of words without deep explanation. for ex : 1) P(x|y) , no mention of what y is. 2) super resolution and signal processing examples : why is this not representation learning , model still learns a representaiton in these cases 3) imitation learning , again why is this not supervised learning (just because its RL) . Too much usage of jargon and no depth

Amazing lectures, thanks a lot! But I am getting a lot of adds on those lectures. For me its around 30 seconds of adds every 5 minutes. It makes it hard to stay locked in... especially on this lecture, which demands a bit more effort from me

Hey Volodymr -- Any chance that additional lectures will be added? I loved this first one

Hi, where can I get the slides of the course?

its only theory it would have been so much better if done practical side by side too.

One of the most amazing lectures. Ive never seen a lecture on generative models that is so connected like these, from simple autoregressive, to latent models, to gans, to energy-based, langevin dynamics, and finally to diffusion models, all are connected! The connectedness and story telling are so amazing! thank you Prof!

very high quality and amazing lectures, thank you Prof!

Prof could you please upload the slides since your github slides arent updated?

Please use generative AI to super resolution the audio behind this video!

Why is pixel x3 dependent on x1, x2, but not x4..? Okay I get it it is using Chain rule in Probability to decompose the joint probability distribution.. Not saying x3 depends only on x1 and x2, but learni learning that marginal distribution as a way to predict joint distribution.

applied machine learning but i can only see ppts not practical implementation? how can this be applied?

I had been looking for CS236 videos and couldn't find any, then the next day I find this course that teaches from the same slides. Woohoo!!!

Thanks a lot. I really liked your teaching method. Some channels do not go into the details of the statistics but statistics is more important than just applying the method.

Great lecture, thanks!

Thank you for releasing this online! You're a legend

Thanks so much for fixing the audio! Looking forward the course!

why is the score function graph at around 21:00 postive after 5ish? As soon as you past the point corresponding to the mode at around 5 shouldn't it point left (be negative)? Same question for the other mode around -4

Thanks for sharing such good materials! Does RNN belong to autoregressive models?

excellent lecture, horrible sound

Thanks for sharing these amazing lectures!

There are too much noise in audio. It will be great if you enhance the audio by using a generative model. Thanks for the video by the way...

Thanks for the lecture!

Hello, Thank you for providing this valuable course on generative modeling. I would like to ask a question about slide 18. Should the multiplication for alpha t hat be from alpha 1 to alpha t (small t) instead of alpha T (capital T)? Is my understanding correct?

Presenting the unified view of Energy, Score and Diffusion models is invaluable. My coursework didnt cover generative methods beyond VAE and GANs but this presentation has been a great surrogate. Thanks! A small correction to the proof on slide 14. I think there's possibly a rogue squared s_theta(x) in the third term.

Can you please share the slides? Thanks. It is an interesting and useful course!

Amazing Lectures! Can't wait to watch the next diffusion lecture! Awesome!

First comment !! I guess watching videos fully has its advantages