Grounded Semantic Simulation using Deep Learning

Description

We simulate future states of game boards when we play games like chess or go, anticipating moves of the opponent and their consequences. There are indications that the human brain simulates the real world too. This enables us to take decisions, consciously or not, to solve concrete problems, to understand language, etc… Starting from existing approaches for computer understanding of complex scenes through deep learning technology, the researcher will investigate new approaches to enable the prediction of future states of such complex scenes, an important capability of a simulator system. The research work is actually related to the topic of grounded computer understanding, which is considered as a necessity to really solve many concrete problems in artificial intelligence: automatic translation, situated human-computer interaction, autonomous robots, etc… The research will allow the candidate to develop a unique and transferable expertise in advanced machine learning, AI and Big Data. The researcher will be integrated within a significant group of deep learning researchers involved in regional and international collaboration with world-leading labs in the area.