Interpretable, scalable and flexible AI for video games

适用于视频游戏的可解释、可扩展且灵活的人工智能

• 批准号: 2441791
• 金额: --
• 依托单位: Queen Mary University of London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2441791
• 关键词: Interpretable; scalable; flexible; AI; video

I propose to introduce Natural Language Processing (NLP) techniques and continue the effort to develop efficient model-based reinforcement learning agents, a topic that is currently investigated at QMUL. Several works have already been published to introduce Natural Language informed Reinforcement Learning agents. For a recent review of the topic see (Luketina et al., 2019) and for a recent application that uses Natural Language to generalise the dynamic of a Reinforcement Learning environment (Zhong, Rocktaeschel, & Grefenstette, 2019). Model-based RL is gaining increasing importance in recent years. One very interesting approach is presented by (Berkenkamp, Turchetta, Schoellig, & Krause, 2017). The authors present an approach that leverages the statistical properties of the dynamics of the environment to ensure that the agent safely collects data. I believe that I can apply this technique extensively in video games AI to model and encourage safe behaviour under uncertainty situations.

我建议介绍自然语言处理（NLP）技术，并继续努力开发有效的基于模型的增强剂学习剂，这一主题目前在QMUL进行了研究。已经发表了几部作品，以介绍自然语言知情的增强学习者。有关该主题的最新综述，请参见（Luketina等，2019），以及最近使用自然语言来推广强化学习环境动态的应用程序（Zhong，Rocktaeschel和Grefenstette，2019年）。近年来，基于模型的RL正在越来越重要。 （Berkenkamp，Turchetta，Schoellig和Krause，2017年）提出了一种非常有趣的方法。作者提出了一种利用环境动力学的统计属性的方法，以确保代理安全收集数据。我相信我可以在视频游戏AI中广泛应用这项技术，以建模并鼓励在不确定性情况下进行安全行为。