Contents

## How does a VAE work?

VAE is a generative model – it estimates the Probability Density Function (PDF) of the training data. If such a model is trained on natural looking images, it should assign a high probability value to an image of a lion. An image of random gibberish on the other hand should be assigned a low probability value.

### Why is autoencoder better than Varienal autoencoder?

A variational autoencoder assumes that the source data has some sort of underlying probability distribution (such as Gaussian) and then attempts to find the parameters of the distribution. Implementing a variational autoencoder is much more challenging than implementing an autoencoder.

#### Are VAEs Bayesian?

Variational autoencoders (VAEs) have become an extremely popular generative model in deep learning. While VAE outputs don’t achieve the same level of prettiness that GANs do, they are theoretically well-motivated by probability theory and Bayes’ rule.

**What’s the difference between a VAE and a sample?**

The two primary differences – that samples are encoded as two vectors that represent a probability distribution over the latent space rather than a point in latent space and that Kullback-Leibler divergence is added to optimization – will be covered in more detail. Through these, we’ll see why VAEs are suitable for generating content.

**What can a variational autoencoder ( VAE ) be used for?**

Like GANs, Variational Autoencoders (VAEs) can be used for this purpose. Being an adaptation of classic autoencoders, which are used for dimensionality reduction and input denoising, VAEs are generative. Unlike the classic ones, with VAEs you can use what they’ve learnt in order to generate new samples.

## Which is the reasoning that leads to Vaes?

Building, step by step, the reasoning that leads to VAEs. This post was co-written with Baptiste Rocca. In the last few years, deep learning based generative models have gained more and more interest due to (and implying) some amazing improvements in the field.

### What are the dimensions of a Vaes image?

The images are 28×28 pixels, so if we add a convolutional layer with a stride of 2 and some extra padding too, we can reduce the image dimension to the half ( review some CNN theory here ). With two of these layers concatenated the final dimensions will be 7×7 (x64 if have into account the number of filters I am applying).