Variational Autoencoders Theory Explained | Generative AI Course

[@jayeshkurdekar126]

The main takeaway is sensitvising decoder to the joint distribution...cool.great video thanks

[@datafuseanalytics]

Thanks a lot brother. I am very happy that you liked it. 👍

[@sekhardhana3453]

Thanks for the video bro, much appreciated

[@datafuseanalytics]

Hey Sekar, Thanks a lot. Do share with other Data Science Enthusiasts... 😃

[@asheeshmathur]

Good Video clarifying VAE

[@datafuseanalytics]

Thanks a lot Asheesh.. glad it helped you.

[@datafuseanalytics]

Hello AI Enthusiasts, In this video I tried to simplify VAE for you all. Do tell me your feedback. 0:00 Introduction to Variational Autoencoders 0:20 Problems with Autoencoder 1:39 Variational Autoencoders to the RESCUE 2:41 What is Mean vector (μ) and Standard deviation (σ)? 3:09 VAE Model Architecture 4:43 How this change in Encoder of VAE helps? 6:27 What we expect vs what we get from VAE? 6:41 Lets welcome a new LOSS function 7:13 Why we need both Reconstruction and KL Loss? 8:20 THE REPARAMETERIZATION TRICK

[@theadchannel1168]

Really good explanation

[@datafuseanalytics]

Thank you. I am glad that you liked the video

[@asheeshmathur]

As usual the best description of VAE, @04:26, you said it develops mean and variance(its log) for each dimension. Does it refers to each channel of image . Also what is need for z_sigma and epsilon. 1. Is Mean and Variance not enough

[@datafuseanalytics]

Hello. In a Variational Autoencoder (VAE), the mean and variance are crucial for defining the parameters of a distribution in the latent space. However, sigma and ϵ (random noise) are essential for the reparameterization trick. This trick enables the network to sample latent representations by decoupling the sampling process from the network's parameters during training. zsigma scales ϵ to match the desired distribution N(μ,σ), making the sampling process differentiable and allowing for effective learning of meaningful latent representations while reconstructing input data.

[@Zyx2003]

Your video is wonderful!You did save my life ,thanks!😀

[@datafuseanalytics]

Hey thanks a lot. I am glad that this video was helpful. 👍 😃

[@pvtgcn8752]

❤

[@datafuseanalytics]

Thanks a lot ❤️

[@flakky626]

Go in more depth lol VAE's are not something you can just learn with 9 minutes video.. Can you recommend best resource to learn VAE from please?

[@mthornit]

Jakup Tomczak's deep generative modelling is pretty good. Also Umar Jamil's VAE video

[@datafuseanalytics]

Hello, if you want to learn in depth, I would refer this beautiful book - "Generative Deep Learning" by David Foster. (www.oreilly.com/library/view/generative-deep-learning/9781492041931/)