An Introduction to Deep Generative Models for Computer Vision

In contrast with discriminative models that are interested in the probability of the world state y given the data x, generative models focus on the opposite, trying to learn the structure of the data depending on the state of the world. This different approach allows us to generate new samples of x. For example, generative models are often used in computer vision to produce realistic but completely artificial pictures.

This meetup will cover two of the most famous and recent generative models, Variational Autoencoders and Generative Adversarial Networks, at an introductory level with the main mathematical insights. The theory behind the techniques and how they work will be presented, together with practical applications and code examples.

The event will consist of two presentations of 45 mins each, with an additional 15 mins for questions and technical time.

• First Talk: Variational Autoencoders (VAEs)

Bio:
Marco Odore, Riccardo Fino and Stefano Samele are data scientists at Agile Lab, with a computer scientist, statistical and mathematical background respectively.

Abstract:
The Variational Autoencoder is a model that maps the input data x in a low-dimensional and more informative latent space z. It then tries to reconstruct the original input from z to verify its accuracy. The VAE introduces a probabilistic interpretation to traditional autoencoders, allowing to estimate uncertainty and also to leverage domain knowledge.
In the example will be shown how to train a VAE on an image dataset and will be given a glimpse of its interpretation through visual arithmetics.

• Second Talk: Generative Adversarial Networks (GANs)

Bio:
Luca Ruzzola is a deep learning researcher at boom.co, working on applied Deep Learning for photo enhancement.

Abstract:
Generative Adversarial Networks (GANs) have revolutionised the field of generative deep learning in the last few years, and have obtained incredible results in various applications. All of this stems from an interesting idea on how to train neural networks, rather than from a new architecture, in contrast with most other innovations in the field. In this talk we're gonna go back through their development, the challenges in using them, and the amazing results that they can produce, from an introductory perspective.

This meetup is 100% COVID-19-free and you will attend it online.
The streaming platform is Teams, the link provided is the direct access for attendees.
Desktop: you can access via browser or with the Teams app.
Mobile: you need to access with the Teams app, unless you set your mobile browser as "Desktop Website".

We hope you will join us!
Please, send your RSVP as soon as possible in order to let us organize the event in the best way possible.

See you all on Thursday 11th June!