Industry Relations: Neural MMO: A Massively Multiagent Game Environment for Reinforcement Learning
10 May 2022 (Tuesday)
Online via Tencent Meeting
Organised by the IEEE Computational Intelligence Society Shenzhen Chapter.
Activity Aim & Talk
In this virtual activity, Jiaxin Chen, a reinforcement learning researcher from Hyperparameter Technology, will introduce some of their applications, explorations and thinking in the field of reinforcement learning and Game AI, and also introduce a massive agent reinforcement learning research environment they are working with, Neural MMO.
Neural MMO simulates the ecological environment and designs a mechanism for resource collection and agents to attack each other. Agents can cooperate with each other or attack each other in the environment. Each map has a minimum of 16 teams, each team has 8 agents. The agents must learn to survive on the map, learn to cooperate with teammates, and master skills such as collecting, exploring, attacking and upgrading equipment. In order to cope with the evaluation method of the 16 models “Brawl Mode”, the agent must learn to face many different strategies that are not visible in training on a random map, and also learn to make decisions based on local observations. Multi-agent, multi-task reinforcement learning, robustness, and generalization, these very fundamental research topics that are well carried out in Neural MMO and are very gamified in a very interesting way.
Meet the Speaker
Chen Jiaxin, graduated from Sun Yat-Sen University in 2016 and the Hong Kong Polytechnic University in 2021, with the main research direction of machine learning theory and algorithms. During her PhD, he has published academic papers in machine learning summits such as NeurIPS, and AAAI, and served as a reviewer for ICML, NeurIPS, ICLR, AAAI, etc. In 2021, she joined Hyperparameter Technology and was responsible for the cutting-edge exploration and research of the direction of Game AI.