To find out how it all began, we have to go back in time to over thirty years ago. It was in 1990. Frédéric Alexandre had recently completed a PhD on the functional modelling of the cortex and joined the SYCO project as a researcher at Inria Nancy. The unique thing about this project was that it was one of the first to work on Artificial Intelligence at Inria, addressing both its digital and symbolic aspects. Ten years later, the researcher decided to develop his research by orientating it toward neuroscience with the creation of the Cortex project, whose main objective was to focus on the neuronal side of artificial intelligence (AI) with an increasing emphasis on the bio-inspired aspects of neuronal computation.
Reproducing brain mechanisms to improve AI algorithms
These two experiences inspired him to concentrate his work more fully on neuroscience in 2012. Based on the observation that the existing artificial intelligence systems, although very efficient in their individual fields of competence, which actually corresponded to basic cognitive functions (recognition, prediction, control), were still a long way off human higher cognitive functions (reasoning, problem solving, creativity etc.), Frédéric Alexandre suggested creating a new project: Mnemosyne.
Our aim is to understand how the brain executes these functions and to propose neuronal models capable of reproducing their key properties, in particular the flexibility of our problem-solving strategies, the acquisition and manipulation of knowledge, its semantic foundation and the explicability that allows us to justify our decisions.
The Mnemosyne project-team manager
Hosted in a neuroscience laboratory on the Bordeaux Neurocampus, the Mnemosyne project team now works hand in hand with different research teams from the Institute of Neurodegenerative Diseases (IMN) on various themes: the modelling of decision making and reward coding, the temporal organisation of behaviour and processing the data generated by the different teams.
“Through our work on bird song and human-robot interaction, we aim to better understand how language and semantics are produced from interactions between the body and the brain. The work carried out in neuroscience on decision making in rodents and primates allows us to produce a general modelling framework to better understand the interactions between implicit and explicit memory, between knowing how to do something and knowing about something”, explains Frédéric Alexandre.
Sharing experience internally and externally
The results of Mnemosyne’s work are intended to be transferred to the fields of neuroscience, medicine, artificial intelligence and machine learning. The project team also shares its experience in artificial intelligence with the general public by taking part in the Class’Code IAI MOOC, for example, which is open to all audiences, and in the annual AI For Industry forum dedicated to developing artificial intelligence for industrial transfer.
Mnemosyne’s work also contributes to the AIDE (Artificial Intelligence Devoted to Education) exploratory action, which aims to analyse the potential impact of approaches or techniques from cognitive neuroscience, in connection with machine learning and symbolic tools to represent knowledge, on human learning as studied in education science.