news

shanghai, september 25 (reporter yang yu) this morning, the "decoding the future: global digital intelligence trends" special forum was successfully held in pudong, shanghai. this special forum is one of the sub-forums of the international forum on industrial civilization, hosted by the industrial culture development center of the ministry of industry and information technology, shanghai municipal economic and information commission, and shanghai pudong new area people's government.

at the forum, mathieu laurière, assistant professor of mathematics and data science at nyu shanghai, shared machine learning and generative ai for large-scale multi-agent systems. multi-agent systems or multi-agent systems are not uncommon in our daily lives, such as crowd motion, traffic routing in large cities, and multi-agent systems are often used in financial markets. the characteristics of multi-agent systems are that there are a large number of agents that interact and make decisions, and the environment is very complex.

mathieu laurière delivered a keynote speech. image source: provided by the organizer

mathieu laurière pointed out that people's behavior and decision-making can be understood through multi-agent systems. for example, if a subway station is crowded, we want to understand the behavior of everyone in the subway station, such as which area has a higher density of people, which area has a lower density, and what their movement trajectory is like. however, as time goes by, the density of the crowd will change, which means that their behavior will change. "we want to understand the principles behind their changes in behavior. in the field of transportation, we can use multi-agent systems to understand the operation of transportation." mathieu laurière said.

what is the connection between multi-agent systems and generative ai? mathieu laurière said that to understand people's actions, that is, the behavior or decision of the agent, it is necessary to generate a distribution map of the agent, which may solve many problems of generative ai. in addition, in terms of deep learning, the deep neural network can be parameterized to be as close to the real situation as possible, and then the data can be mapped onto it.

daily economic news