上海财经大学 > 科学研究 > 学术交流 > 学术报告
  • 统计与管理学院2017年学术报告第84期

    【主 题】 Toward an Understanding of Deep Learning for Data Analysis: the Expand-and-Reduce Method

    【报告人】 Chuanhai Liu 教授

    Purdue University

    【时 间】 2017年12月27日(星期三)10:00-11:00

    【地 点】 上海财经大学统计与管理学院大楼1208会议室

    摘 要】Big data analysis demands computer-intensive data-driven methods more than ever for exploring and modeling complex data structures. Because of its capability and flexibility to represent complex local and global data structures, ``multi-level nets'' or Deep Learning models can be practically very useful. However, architecture configuration and parameter optimization of such models are extremely challenging, both statistically and computationally. To overcome the difficulty, we propose an Expand-and-Reduce method for automated building of Deep-Learning models. The purpose of the method is three-fold: 1) it generates models for Artificial Intelligence-type of applications, 2) it can be used to do confirmatory-type analysis for investigating prior knowledge-based construction of network structures, and 3) it provides as a tool for in-depth investigation and understanding of data from scientific inference perspective. With simple examples, we show that our proposed method is promising to serve its purpose.