Deep Learning : Je t'aime moi non plus..
Jean-Luc Parouty · HAL (Le Centre pour la Communication Scientifique Directe) · 2019
Reproducing or simulating **"intelligence"** is an objective that has long occupied us and whose first scientific works are older than computers.Long overlooked, **machine learning** techniques and more particularly **deep learning**, based on **neural networks**, have made exceptional progress in recent years.Carried by the major Internet players, these technologies have become essential in the **"big data"** revolution.By linking experimental and modelling and opening up new approaches in the **scientific method**, these technologies have considerable potential in most of our scientific fields.The objective of this presentation is to provide simple and synthetic **elements of understanding** on what **deep learning** is, its principles and uses.We will approach this presentation as a triptych: * The first part will tell the tumultuous and uncertain story of these **artificial neurons** from 1940 to the present day, * We will then explore the different **architectures** and their possible **uses**, based on short demonstrations, * We will finish (probably late) on the **service offers** and **future prospects**, especially regarding our infrastructures and businesses.The whole presentation will be accessible to our natural intelligences 3.0 and +Many thanks to www.DeepL.com/Translator and his artificial neurons for this translation !