Unsupervised conceptual extraction in Deep Neural Networks. Stemming from the recent advances in Explainable AI (XAI), the work will train deep neural networks to classify well-known paintings according to standard artistic taxonomy (genre, style, ...) to ultimately generate and study the "artificial concepts" that led to a decision. Such "artificial concepts", representing high-level entities such as use of color and strokes, will be autonomously uncovered by the network, thanks to Logic Layers, and later analyzed by an expert.