统计与数据科学学院
-------------------------------------------------
学术报告
报告题目：Efficient Online Learning in Multi-dimensional Decision Problems
讲座人：郝博韬博士
普渡大学
时间：12月28日上午10:00-11:00
地点：统计与数据科学学院426教室
摘要：Traditional static multi-dimensional recommendation system assumes the users pref-erence over the item, location and device does not change over time. In many recommendation domains such as email campaign, users constantly interact with the systemwith dynamic preference, and user feedback is instantly collected for improving recommendation performance. In these settings, it is essential for the recommendationmethod to adapt to the shifting preference patterns of the users. In this talk, we consider a stochastic low-rank tensor bandit which naturally accounts this dynamics,and propose a two-stage online tensor estimation method for interactive recommendationwith multi-dimensional actions. In theory, the cumulative regret of our model is studied and compared to the corresponding vectorized version to justify the importance of low-rankness consideration in tensor bandit. The superior performance of our proposed method is illustrated on the online recommendation task of Adobe email marketing campaign.
邀请人：邹长亮教授
欢迎广大师生参加！
统计与数据科学学院
2018年12月28日