100% Agree that if you write everyting from scratch line by line it is much better than having it prewritten

Great content. I've always loved your "from scratch" tutorials.

@AladdinPersson

Жыл бұрын

Thanks Marco!

You are truly a life saver sir. Thank you for keeping everything simple instead of using programming shenanigans just to make it more complicated and unreadable. Love your tutorials, I learned a lot from your line of thinking, including the ranting things.

Hey just watched your video, really good! But its obvious this is a new area for you (which is not bad), so I thought I'd give you some pointers to improve your algorithm. 1. In practice VAE's are typically trained by estimating the log variance not the std, this is for numerical stability and improves convergence of the results so your loss would go from: `- torch.sum(1 + torch.log(sigma.pow(2)) - mu.pow(2) - sigma.pow(2))` -> `-0.5 * torch.sum(1 + log_var - mu.pow(2) - log_var.exp()` (where log_var is the output of your encoder, also your missing a factor 0.5 for the numerically stable ELBO) Also, the ELBO is the Expectation of the reconstruction loss (the mean in this case) and the negative sum of the KL divergence 2. The ELBO (the loss) is based on a variational lower bound its not just a 2 losses stuck together as such arbitrarily weighting the reconstruction loss and the KL divergence will give you unstable results, that being said your intuition was on the right path. VAEs are getting long in the tooth now and there are heavily improve versions that focus specifically on "explainable" if you want to understand them I would look at the Beta-VAE paper (which weights the KL divergence) then look into Disentagled VAE (see: "Structured Disentangled Representations", "Disentangling by Factorising") these methodologies force each "factor" into a normal Gaussian distribution rather than mixing the latent variables. The result would be for the MNIST with a z dim of 10 each factor representing theoretically a variation of each number so sampling from each factor will give you "explainable" generations. 3. Finally your reconstruction loss should be coupled with your epsilon (your variational prior), typically (with some huge simplifications) MSE => epsilon ~ Gaussian Distribution, BCE => epislon ~ Bernoulli distribution

Great video, thank you! Please don't change to having pre-written code. Your approach is the best that can be found these days.

As usual, it's amazing content ! Thank you so much for your work

I like the thought process. So, thanks for the 'from scratch' tutorials.

I prefer from scratch too for all the reasons you've mentioned. Thanks for the content .

Awesome work. Please do more stuff with GANs or visual transformers.

Thanks for your amazing implemention and interpretation!

I love "from scratch" series, plz make more videos..!! and thank you so much!!!

Thank you very much for your tutorials. They have been incredibly helpful and insightful.

Again an awesome from-scratch video! I have never seen programming videos in which it is so simple to follow what the person is coding, thank you. Currently, there are no videos about stable diffusion from scratch, which include the training scripts. It would be great to see a video on this!

Your videos are shockingly good! Among programming channels it is the best one.

@AladdinPersson

Жыл бұрын

Appreciate you saying that

very informative love the explanation of content and implementation from scratch

Awesome Dude!!!! So great!!

Really helpful! you are awesome!!!

Thanks Aladdin, you helped me a lot, thanks for the unique explanation, keep up the good!

Awesome implementation tutorial❤️

you are awesome. thank you for this immensely valuable resource!!

Finally kicking off this series, I've been waiting for years. Curious if you'll do VQ-VAEs like in the Jukebox example from OpenAI?

@AladdinPersson

Жыл бұрын

Yeah.. Don't have a structured plan for what's next but VQ-VAEs would be cool to understand

great explanation, thanks!

Very informative content . Also can you make shorts that explains small stuffs

@AladdinPersson

Жыл бұрын

Thanks Dhairya! Good idea, haven't figured out what to make on yet but will think about it:)

23:09 since you used sigmoid your pixels will be between 0 and 1 so it's okay to use sigmoid in this case otherwise if you use no activation function in the last layer of the decoder you need to use the new loss function of MSE +Reconstartion loss that what i think

lovely video man, thankyou

Dope stuff!

Thanks for the tutorial, it was simple yet insightful. Can you also make a video where you can combine different architecture such as Transformers or Residual blocks in Encoder-Decoder block of VAE.

Had to learn about VAE with zero experience in coding or ML. Thank God I found this video 😅

So fast！Awesome！

I see few videos of you about GAN, so probably you want to have a look at Adversarial Autoencoders. Instead of using KLD, you can impose a prior on the latent using a discriminator.

Great tutorials!! I can understand how to work on VAE!! ☺☺☺☺

Amazing!

ALADDIN PERSSON. YOUR CONTENT IS AMAAAZZZIIINNGGG !!! THANK YOU FOR PRACTICAL DEEP LEARNING WITH PYTORCH

@AladdinPersson

Жыл бұрын

Thanks & np!!

@teetanrobotics5363

Жыл бұрын

@@AladdinPersson If possible, please include this in the playlist and make more tutorials please. Loving it !!

Maybe the "Attention Is All You Need" is worth to go through

Thanks a lot, is there any recommendation on TensorFlow VAE tutorial ?

@AladdinPersson

Жыл бұрын

Recommendation is to use pytorch ;)

I'm speechleess, the content is too good

Miss it!

First of all, thank you very much... Secondly, in line 74, should'nt we have epsilon = torch.randn_like(1) instead of epsilon = torch.randn_like(sigma)? Because we want an epsilon distributed in N(0,1) and then the next line will generate z which will be distributed in N(sigma, epsilon).

Doing it from scratch is way better than just typing some pre-written code.

@AladdinPersson

Жыл бұрын

What do you mean by this?

do more vids about vision transformers

PyTorch has a loss function for KL divergence, I was wondering if it's possible to use that instead of writing it?

@AladdinPersson

Жыл бұрын

Yeah that should be possible.. haven’t tried it though.

@AladdinPersson

Жыл бұрын

Yeah that should be possible.. haven’t tried it though.

Code from 15:05 so you don't need to type it all: import torch import torchvision.datasets as datasets from tqdm import tqdm from torch import nn, optim from model import VariationalAutoEncoder from torchvision import transforms from torchvision.utils import save_image from torch.utils.data import DataLoader

isnt self.activation = nn.relu better?

@AladdinPersson

Жыл бұрын

Yeah, maybe slightly confusing if we’d be using multiple activations?

@fizipcfx

Жыл бұрын

@Aladdin Persson i guess you are right that way its more clear

I just have to say that, even as someone with a Master's in Data Science from a top university, I still use your tutorials for my work and my projects. Your stuff is incredibly helpful from a practical perspective. In school, they teach you theory with little to no instruction on how to actually build anything. Thank you so much for your hard work!!

why machine learning is easy to learn? Because a lot of amazing guys are making videos about explaining papers and writing codes line by line.

@AladdinPersson

Жыл бұрын

Thanks for the kind words ❤️

are you the son of notch (markus persson)?

please do not have the code prewritten

@AladdinPersson

Жыл бұрын

Agree. I get overwhelmed if someone shows the entire code. Much easier to get guided through it step by step imo, but open to the idea that there might be better ways to explain code

Its always to good write the code from scratch....

Merci !

mnist dataset lol. all samples/videos using the same DS. so boring. create your own dataset, implement something interesting