香港科大 杨强老师的PPT 关于迁移学习的基础理论和应用 适合初学者机器学习:昨天,今天,明天昨天今天明天深度学习强化学习:迁移学习大量数据大量数据少量数据特征反馈学习方法准确性策略冷启动/个性化Transfer LearningIn the machine learning communitylearned in previous domains/tasks to novel tasks domains, whichThe ability of a system to recognize and apply knowledge and skillshare some commonalityGiven a target domain/task, how to transfer knowledge to newdomains/tasks(target)?Representation Learning, Change of Representation02迁移学习y iransferBuild every model from scratch?日 Expense on Data:CollectionLabelingTrainingCold Start ProblemRecommendation Systems, Learning SystemsPersonalization The next Al frontierU User Privacy Problems, Cloud to Client Transfer (A到端传输)迁移学习的好处?学习效果P曲线一:使用迁移学数据一定P1效果:Cioa0ostpnoP1>P2曲线二:P2不使用迁移学效果一定数据:99D1