Generative Modeling of Convolutional Neural Networks

• Activity Number: 514
• Type: Invited
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: General Methodology
• Author(s): Ying Nian Wu*
• Companies: UCLA
• Keywords: Big data ; Deep learning ; Generative pre-training ; Visualization of ConvNet

Abstract

The convolutional neural networks (CNNs) have proven to be a powerful tool for discriminative learning from big data. Recently researchers have also started to show interest in the generative aspects of CNNs in order to gain a deeper understanding of what they have learned and how to further improve them. In this talk, I will present our recent work on generative modeling of CNNs. The main contributions include: (1) We construct a generative model for CNNs in the form of exponential tilting of a reference distribution. (2) We propose a generative gradient for pre-training CNNs by a non-parametric importance sampling scheme, which is fundamentally different from the commonly used discriminative gradient, and yet has the same computational architecture and cost as the latter. (3) We propose a generative visualization method for the CNNs by sampling from an explicit parametric image distribution. This talk is based on the joint work with Jifeng Dai and Yang Lu.