Thesis of Jianyong Xue
How can an agent build knowledge of the environment and of itself effectively and efficiently without any predefined information? In our research, we are interested in investigating the issues as following: i) Put an agent in an unknown environment without any predefined information, let the agent construct spatiotemporal knowledge based on its interactions with the environment. ii) Create a cognitive architecture that can let the agent discover and exploit information from the traces that the agent produces when interacting with the environment. This research falls within the scope of unsupervised and self-motivated learning as opposed to reinforcement learning in which the experimenter defines the agent's reward.
Through our research, we hope to build a more powerful cognitive architecture that can meet the needs of a single-agent and or of a multi-agent system to build knowledge about the environment and itself and to revise its behavior more quickly and efficiently by interacting with the environment to adapt to it.
Advisor: Salima Hassas
Coadvisor: Olivier Georgeon