My project (#11) is about capturing and computationally reproducing some of the essential characteristics of creative cognition and insight. More precisely, I am interested in "thinking outside the box": the ability to depart radically from a tentative strategy in order to try something new. Developments in Reinforcement Learning, especially with regards to temporal abstraction, have brought exciting new tools to tackle this problem at a fundamental level of cognition.
The agents developed as part of this project learn policies (strategies), structured hierarchically or making use of what I call "intentions", which can direct their behaviour at different levels of abstraction. Such agents learn not only by discovering a new sequence of elementary actions, but also by deciding between a large range of abstract vector-valued strategies – from the "big picture" to precise motor control. This is expected to greatly reduce costly trial-and-error exploration, and to allow for quicker adjustment to new environments and discovery of radically novel solutions to a problem.
Biological agents with enough experience are adept at making radical changes between abstract policies – sometimes experiencing the illumination of a eureka-moment when the new big picture seems to hold unexpected promise. This project makes use of the considerable literature in psychology and neuroscience in order to develop agents inspired by biological creativity.
I hold master’s degrees in Software Engineering (EFREI, France, 2008) and Cognitive Artificial Intelligence (Cum Laude, Utrecht University, The Netherlands, 2013). I have previously worked in Cognitive Modeling in the context of Urban Search and Rescue robots, with TNO Soesterberg and as part of the E.U. project "NIFTi".
I also have experience as a software engineer (2008-2010, 2013-2014), and I have studied philosophy and cognitive psychology at Université Toulouse II-Le Mirail.
