The MMD-VAE modifies the standard VAE by replacing the KL-divergence term in the loss function with a Maximum-Mean Discrepancy (MMD) term, which measures the distance between the aggregated posterior of the latent codes and the prior.

In machine learning, a variational autoencoder (VAE) is an artificial neural network architecture introduced by Diederik P. Kingma and Max Welling. [1] It is part of the families of probabilistic graphical models and variational Bayesian methods. [2]

Jun 10, 2017 · This is a PyTorch implementation of the MMD-VAE, an Information-Maximizing Variational Autoencoder (InfoVAE). It is based off of the TensorFlow implementation published by the author of the original InfoVAE paper.

This tutorial discusses MMD variational autoencoders (MMD-VAE in short), a member of the InfoVAE family. It is an alternative to traditional variational autoencoders that is fast to train, stable, easy to implement, and leads to improved unsupervised feature learning.

The paper Auto-Encoding Variational Bayes combines variational inference with autoencoders, forming a family of generative models that learn the intractable posterior distribution of a continuous latent variable for each sample in the dataset. This repository provides an implementation of such an algorithm, along with a comprehensive explanation.

This paper proposes a new objective function for variational auto-encoder, a widely used image generation architecture in data augmentation. By replacing KL divergence with Wasserstein distance to form a new variational lower bound, we show that ELBOW has a better theoretical approximation and more efficient

In this paper we combine these ideas to build a variational ladder autoencoder with MMD loss instead of KL divergence, and utilize this model to analyze of structure and hidden features of human faces. As an application we use this model to perform “arithmetic” operations on faces.

Oct 21, 2018 · Like GANs, variational autoencoders (VAEs) are often used to generate images. However, VAEs add an additional promise: namely, to model an underlying latent space. Here, we first look at a typical implementation that maximizes the evidence lower bound.

For a generative model of graph statistics, we adapt calibrated Gaussian variational auto-encoder [41], which has been developed as a generative model for i.i.d. data, but not previously been applied to graph modeling.

Implementation of the MMD VAE paper (InfoVAE: Information Maximizing Variational Autoencoders) in pytorch