Machine Learning

机器学习

• 批准号: 1000231167-2015
• 负责人: Precup, Doina
• 金额: $ 7.29万
• 依托单位: McGill University
• 依托单位国家: 加拿大
• 项目类别: Canada Research Chairs
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=642723
• 关键词: Machine; Learning

Automated intelligent agents are everywhere, from voice recognition on our phones to sophisticated clinical monitors. Such agents need to adapt to unpredictable situations, without external intervention. Reinforcement learning describes how they can be trained, using numerical rewards. But how quickly should such an agent act and model the world? For example, a driving agent may need both a long-term model of traffic, in order to plan a route, and a short-term model of obstacles, in order to avoid them. This research program provide a solution by taking into account ``thinking" costs. This will allow the development of better algorithms for applications from medical monitoring to activity recognition.

自动化智能代理无处不在，从我们手机上的语音识别到复杂的临床监视器。这些代理人需要在没有外部干预的情况下适应不可预测的情况。强化学习描述了如何使用数字奖励来训练它们。但是，这样一个代理人应该多快采取行动并为世界建模？例如，驾驶代理可能需要长期的交通模型，以便规划路线，以及短期的障碍物模型，以便避免它们。这项研究计划通过考虑"思考”成本提供了一个解决方案。这将允许开发更好的算法，用于从医疗监测到活动识别的应用。