To end this series of articles on the AI & Art project I realized in collaboration with Cali Rezo, today I would like to give you some of my thoughts on this project and on AI in general, plus some info on the tools we used and on Cali’s upcoming events.
Over the past few weeks, we explored several questions related to AI and its application to art generation or analysis. Before finishing the series, let’s look at things from Cali’s point of view and talk a bit about artists and technology…
In the third article of the series, we discussed how to apply AI to art analysis. Even if our results were not as conclusive as we’d hoped, they still raised a few questions that we will tackle today: what is really happening in these black box models that are neural networks? And to which extent can we assess how certain a model is of its predictions?
In the last article, we focused on VAEs and GANs for image generation. This time, we’ll talk about analyzing images and trying to identify classes. We will also take this opportunity to talk about the usual traps and limits of AI classification.
To start off with this series of articles on the AI & Art project I did in collaboration with Cali Rezo, we’ll discuss some common generative models and how we applied them to her artwork to create new images in a “Cali-like” style.
Nowadays, machine learning (ML) is a red-hot topic which is discussed everywhere in various ways. More and more companies are relying on AI as part of their production process, be it in the domain of finance, medicine, management, art… This last application of ML algorithms, in particular, is really interesting to me. And, since the great abstract painter Cali Rezo shares this interest, we decided we would collaborate on a project to study how to apply AI to art.
In this last article from the FEM series, I present some additional concepts related to the finite element method and further developments. In particular, I focus on finite element analysis, some methods for solving time-dependent problems and optimization ideas.
Today, let’s continue with the finite element method and focus on how to implement it on a computer. How do we discretize our domain? How do we actually compute the stuff that’s in our variational formulation – efficiently, that is? How do we store and visualize our solution?
The study of partial differential equations is a fascinating field of mathematics with many concrete applications, be it in physics, mechanics, meteorology, medicine, urbanism… Mathematicians model the world as equations to better understand it and, if possible, compute a theoretical solution to a physical problem. However, we can’t always get an analytical solution – for example if the geometry is too complex – and sometimes must rely on numerical methods to get an approximation.
Last summer, I did a 2-month internship for the French startup HERETIC which has been fighting scams on the Internet for several years now. They offered me an opportunity to test my AI-engineer skills on a practical problem: how can we use machine learning to detect how fraudulent an email or a website address looks?
Last September, I had the chance to participate in an interview for the “Guide de l’Ingénieur” (i.e.: the Engineer’s guide), a special edition of the French magazine L’Usine Nouvelle that focuses on science and technology. Thanks to the journalist Christophe Bys, 3 other engineering students and I were offered the opportunity to meet with the leader of BCG Gamma, Sylvain Duranton, and ask him a few questions.