Mostafa Rizk: "Teaching Autonomous Agents to Work Together"
Autonomous agents are quickly becoming pervasive in our world, in the form of self-driving cars, warehouse robots, voice assistants and more. As this trend continues, these agents will begin to encounter situations where they can accomplish their goals more efficiently by working together toward a mutual benefit i.e. cooperating. Learning to work together is, however, not trivial. This talk is about understanding the best ways agents can learn cooperative policies. In particular, it is an overview of my PhD research about the trade-offs between centralised and decentralised methods of training agents when the task involves complex social dynamics. It touches on multi-agent systems, neuroevolution, evolutionary game theory and the emerging field of Cooperative AI.
Mostafa is an Egyptian-Australian computer scientist with an international upbringing who moved to Australia shortly after the Arab Spring. He recently completed his PhD in AI from Monash University focusing on cooperative multi-agent learning. Before that he did his Honours thesis on human interaction with robot swarms. He is currently figuring out his next chapter while continuing to expand his knowledge of AI and ML and how they are used in industry. He is interested in neuroevolution, animal intelligence and Cooperative AI as well as all flavours of Deep Learning. When not doing AI stuff, you can find Mostafa salsa dancing, learning to do a handstand or trying to befriend the neighbourhood crows.
Recorded at the Melbourne Machine Learning and AI Meetup in October 2022
mlai.melbourne/
29 авг 2024
