The objective of this course is to present and discuss artificial neural networks and their applications in the study of cognition. The course aims at showing the latest advances in machine learning and its use to understand our cognition, while also highlighting the limits posed by current techniques if considered as models of the brain. In a first part, the fundamental bases, the history and the development of these techniques will be presented. After a presentation of the perceptron, the gradient descent and backpropagation algorithms, the course will present the main contemporary neural methods and architectures, in particular convolutional networks, recurrent networks (such as LSTM), and Transformers, which are modern networks based on a concept of attention that are used in large language models such as ChatGPT. In a second part, guest lecturers will present the use made of these models in their research in cognitive science. Particular focus will be on the comparison between patterns of activations in artificial and biological neural systems. One session will be devoted to the implementation of a learning algorithm. The last part will be devoted to students' presentations, considered as part of the course.

Notice: This course is NOT a substitute for an advanced course on deep learning or more broadly machine learning. This course does NOT address issues in statistical inference.

Recommended Level: Biology, Cognitive Science: M2; Mathematics, Physics, Computer Science: M1 or higher. The course is also of interest for PhD students or postdocs in Cognitive Science or Computational Neuroscience. 

Prerequisites: Elementary mathematics (calculus, linear algebra); Python basics for the TP 
ECTS : 4

Webpage: https://l-bg.github.io/dlcs/