説明
The 58th Seminar
Date and Time: March 3rd 11:00 am – 12:00 pm(JST)
Venue: Zoom webinar
Language: English
Speaker: Stephan Zheng (Salesforce Research)
Title: Economic Modeling, Decision-Making, and Mechanism Design using the AI Economist
Short Abstract:
Solving global socioeconomic challenges, e.g., economic inequality or sustainability, requires new tools and data-driven simulations. To this end, the AI Economist is a multi-agent reinforcement learning (RL) framework that outperforms and overcomes key limitations of traditional economics. I will survey key results in this area: 1) AI tax policies that significantly improve equality and productivity, 2) AI redistribution mechanisms that are preferred over classic ones by real people, 3) AI-powered macroeconomic simulations and boundedly rational AI agents, 4) WarpDrive, our open-source framework for multi-agent RL end-to-end on GPUs, and 5) AI for Global Climate Cooperation, a competition to design climate negotiation and agreements that incentivize global cooperation. Finally, I will survey future research directions towards real-world scale.
Bio:
Stephan Zheng (http://www.stephanzheng.com) is an AI researcher who has been developing the AI Economist, a multi-agent RL framework for economics. Most recently, he led a research team at Salesforce Research. Previously, he also worked on imitation learning, modeling cooperation, and robustness in deep learning and multi-agent games. His research has been published in leading machine learning conferences and scientific journals, including Science Advances, and has led to multiple patents. It has also been widely covered in US and international media, e.g., the Financial Times, MIT Tech Review, Forbes, Volkskrant, podcasts, and Dutch radio. He has served as an area chair for NeurIPS and ICML. Before machine learning, Stephan studied math and theoretical physics at Utrecht University, Harvard University, and the University of Cambridge, receiving the Lorenz graduation prize for his thesis on exotic dualities in topological string theory. He then switched to machine learning research during his PhD at Caltech. He is originally from the Netherlands.
