Variational AutoEncoders (VAE) Implementation

Опубликовано: 04 Август 2026
на канале: Priyam Mazumdar
3,217
85

Code: https://github.com/priyammaz/PyTorch-...

Today we will implement the Variational AutoEncoder! Previously we derived the loss function for the VAE here    • Mathing the Variational AutoEncoder: Deriv...  

This is pretty similar to the basic AutoEncoder, just some new tricks to get it all to work! I think VAEs are some of the coolest things about Neural Networks, but they have some challenges and limitations that I want to explore today!

Timestamps:
00:00:00 Introduction
00:01:00 AutoEncoders vs Variational AutoEncoders
00:02:45 How do VAEs map to Gaussian?
00:09:15 Reparamaterization Trick
00:29:30 LogVariance
00:31:16 Linear VAE
00:45:20 Writing the VAE Loss Function
00:58:20 Training a VAE
01:00:40 Comparing KL Weights
01:02:15 Generating New Samples
01:10:54 Convolutional VAE
01:26:44 ConvVAE Results
01:31:45 Perceptual Loss Functions
01:33:33 Recap

Socials!
X   / data_adventurer  
Instagram   / nixielights  
Linkedin   / priyammaz  
Discord   / discord  
🚀 Github: https://github.com/priyammaz
🌐 Website: https://www.priyammazumdar.com/