Generative Modeling of Convolutional Neural Networks

TL;DR

This paper introduces a novel generative modeling approach for CNNs, including a new pre-training method and a visualization technique, which enhance understanding and performance of CNNs on large-scale image datasets.

Contribution

It presents a new generative model for CNNs, a generative gradient for pre-training, and a sampling-based visualization method, advancing the interpretability and training of CNNs.

Findings

Generative gradient pre-training improves CNN performance on ImageNet.

Sampling-based visualization produces meaningful synthetic images.

The methods are computationally comparable to traditional discriminative approaches.

Abstract

The convolutional neural networks (CNNs) have proven to be a powerful tool for discriminative learning. 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. This paper investigates generative modeling of CNNs. The main contributions include: (1) We construct a generative model for the CNN 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. The proposed visualization method can directly draw synthetic samples for any given node in a trained CNN by the Hamiltonian Monte Carlo (HMC) algorithm, without resorting to any extra hold-out images. Experiments on the challenging ImageNet benchmark show that the proposed generative gradient pre-training consistently helps improve the performances of CNNs, and the proposed generative visualization method generates meaningful and varied samples of synthetic images from a large-scale deep CNN.