Reinforcement Learning for Sports Analytics

体育分析的强化学习

• 批准号: 521357-2018
• 负责人: Poupart, Pascal
• 金额: $ 14.53万
• 依托单位: University of Waterloo
• 依托单位国家: 加拿大
• 项目类别: Strategic Projects - Group
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=694176
• 关键词: Reinforcement; Learning; Sports; Analytics

Our project develops novel machine learning algorithms for interpreting complex, multi-agent scenarios. We will advance the state of the art in deep reinforcement learning methods to enable analysis of continuous-flow input data. The collaboration with our industrial partner will tackle open problems in deep reinforcement learning that can result in richer capabilities in sports analytics for ice hockey and other continuous-flow sports such as basketball and soccer. Deep reinforcement learning, is a breakthrough technology with prominent successes in games such as Go (AlphaGo) and Chess (AlphaZero). This project will push deep reinforcement learning further into the physical world, with multi-person, complex scenarios arising from real-world noisy data sources. We will develop fundamental algorithmic advances and apply them to tasks including: - player evaluation - event predictions (match outcomes, next action, expected scores) - recognizing types of players, teams, play sequences, and tactics - identifying characteristic strengths and weaknesses of players and teams. Our partner is the Montreal-based company SPORTLOGiQ, which uses advanced computer vision to extract information about events from video of sports matches. Their information is more detailed than that provided by any other company or organization. The market for sports analytics is growing rapidly, with major international companies receiving millions of investment dollars. This project will build significant Canadian capacity in sports analytics, support academic research, advance commercialization in the sports industry, and train highly qualified personnel. Canada has achieved a position of leadership in reinforcement learning, through excellent researchers who have attracted companies such as Google's DeepMind to set up Canadian labs. The proposed research will contribute to Canada's reinforcement learning ecosystem by establishing novel algorithmic contributions for a major application area with great commercial potential.

我们的项目开发了用于解释复杂的多代理场景的新型机器学习算法。 我们将在深度加强学习方法中推进艺术的状态，以启用连续流量输入数据的分析。 与我们的工业合作伙伴的合作将解决深度强化学习方面的开放问题，这可能会导致冰球运动和其他连续流动运动（例如篮球和足球）的体育分析能力更丰富。 深度强化学习是一项突破性的技术，在GO（Alphago）和Chess（Alphazero）等游戏中取得了杰出的成功。该项目将进一步将深厚的增强学习进入物理世界，由现实世界中的嘈杂数据源引起的多人，复杂的方案。 我们将开发基本的算法进步，并将其应用于： - 球员评估 - 事件预测（匹配结果，下一个动作，预期得分） - 识别球员，团队，比赛序列和策略的类型 - 确定球员和团队的特征优势和弱点。我们的合作伙伴是位于蒙特利尔的公司Sportlogiq，它使用高级计算机视觉来从体育比赛视频中提取有关事件的信息。他们的信息比任何其他公司或组织提供的信息都更详细。体育分析市场正在迅速增长，主要国际公司获得了数百万美元的投资。该项目将在体育分析，支持学术研究，提高体育行业的商业化以及培训高素质的人员方面建立重要的能力。加拿大通过吸引了Google DeepMind等公司建立加拿大实验室的公司的优秀研究人员，在加强学习方面取得了领导地位。拟议的研究将通过为具有巨大商业潜力的主要应用领域建立新颖的算法贡献来为加拿大的强化学习生态系统做出贡献。