Stein Neural Sampler

• Activity Number: 141 - Statistical Understanding of Deep Learning
• Type: Invited
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Guang Cheng* and Tianyang Hu and Zixiang Chen and Hanxi Sun and Jincheng Bai and Mao Ye
• Companies: Purdue Statistics and Purdue Statistics and Tsinghua Statistics and Purdue Statistics and Purdue Statistics and Purdue Statistics
• Keywords: deep learning; stein discrepancy; neural network; sampling

Abstract

We propose two sampling procedures to produce high-quality samples from a given (un-normalized) probability density. The sampling is achieved by transforming a reference distribution to the target distribution with neural networks, which are trained separately by minimizing two kinds of Stein Discrepancies. Hence, our method is named as Stein neural sampler. Theoretical and empirical results suggest that, compared with traditional sampling schemes, our samplers experience less convergence issues and generate samples instantaneously with statistical guarantee.