Lecture

Which neural code for mental representations ?

from to
Brain endocasts of Australopithecus africanus (purple; over 3 million years old), Homo erectus (specimen KNM-ER 42700, green, approx. 1.55 million years old), Homo naledi (turquoise, approx. 300 000 years old), and a modern Homo sapiens (yellow). - Image by Zachary Cofran, researcher at Vassar College in New York State.

How do the discharges of an assembly of neurons manage to encode each and every one of our thoughts and mental representations, conscious and unconscious ? While the nature of the neural code remains a largely unresolved issue, the perspective on this problem has recently evolved considerably. Moving away from a narrowly cellular vision, centered on the single neuron and the intensity of its response to a variety of stimuli, cognitive neuroscience has now moved towards a geometric, vector-based vision, according to which the neural code is distributed to a whole population of neurons and is therefore implemented by a vector in a very high-dimensional space. This theoretical perspective opens up questions such as the dimensionality and orthogonality of mental representations, their projection onto a subspace, their transformation by mental rotations, and so on. This vision also offers new solutions to the problem of communication between brain areas. But how can we really encode actions, sensations, objects, faces, words, concepts or sentences using vectors ? Can we speak of a geometry of thought ? These are just some of the questions that will be addressed in the lecture.

The lecture will be given in French, with the slides in English.

Program