Vector Quantized Variational AutoEncoder (VQVAE) From Scratch

Code: https://github.com/priyammaz/PyTorch-Adventures/blob/main/PyTorch%20for%20Generation/AutoEncoders/Intro%20to%20AutoEncoders/Vector_Quantized_Variational_AutoEncoders.ipynb To continue on our AutoEncoder adventure, we move onto the VQVAE! We wont spend too much time here training the models, we will just look at the results, as the training code is basically identical to everything we did earlier in our AutoEncoder Tutorial https://youtu.be/p7yUWIySj8o and our VAE tutorial https://youtu.be/9NgC0sh9Msc. Quantization is a powerful tool, especially leveraged in Neural Speech processing and Generative Models, so I wanted to give an introduction here! Its all about keeping backprop alive! Timestamps: 00:00:00 Introduction 00:01:06 KMeans 00:05:10 Review AutoEncoders 00:07:00 What is VQVAE? 00:16:00 Visualize Broken Backprop 00:22:40 Straight Through Gradients Estimator 00:28:00 Visualize Straight Through Estimator 00:31:19 Wheres the VAE? Derive ELBO for VQVAE 00:38:25 Codebook + Commitment Loss 00:43:00 Implement the Vector Quantizer 01:08:00 Implement the LinearVQVAE 01:19:00 Plotting the Embeddings 01:22:40 Implement a ConvVQVAE 01:35:40 Recap Socials! X https://twitter.com/data_adventurer Instagram https://www.instagram.com/nixielights/ Linkedin https://www.linkedin.com/in/priyammaz/ 🚀 Github: https://github.com/priyammaz 🌐 Website: https://www.priyammazumdar.com/