Mehdi Khamassi

WORK HISTORY

EDUCATION

SKILLS

ABOUT MEHDI KHAMASSI

My work is at the interface between cognitive sciences / psychology (understanding the human mind), neuroscience (understanding how the brain works), artificial intelligence (designing algorithms enabling agents to make sense of their perception, to act and to learn), and robotics (designing bio-inspired robots that can interact more naturally with humans, especially for healthcare applications).The goal of my research is twofold:(1) To better understand how decision making and reinforcement learning processes are organized in the mammals\' brain: What are the underlying neural mechanisms in the prefrontal cortex, basal ganglia, hippocampus, and dopamine system? How do they enable humans to adapt so flexibly to new situations? Why and how are they impaired in some neurodegenerative diseases or some psychiatric conditions?(2) To take inspiration from biology to improve current robots\' flexibility and autonomy in decision-making. Among our current healthcare applications, we use small social robots as assistive tools for therapies with children with autism, where the robot is playful and interactive, permitting to better engage the child in the therapy and to mediate and encourage his/her interactions with other children.One of our current central research questions of interest is whether similar learning mechanisms and similar reward processing principles apply to both social and non-social contexts. This is key on the one hand to better understand what is so special about the social dimension of learning mechanisms in the brain, and on the other hand to establish more adaptive and efficient human-robot interactions.Keywords: reinforcement learning; decision-making; set-shifting; auto-evaluation; structure learning; navigation; prefrontal cortex; basal ganglia; dopamine; hippocampus; machine learning; computational neuroscience; autonomous robotics; social robotics; autism; cognitive architectures; artificial intelligence.More info, code and resources: All the pdfs of our scientific articles, open source code of our algorithms, teaching material, and other resources are freely accessible from my professional webpage: https://pages2.isir.upmc.fr/mkhamassi/.