{
  "id": "1412.6296",
  "version": "v1",
  "published": "2014-12-19T11:34:37.000Z",
  "updated": "2014-12-19T11:34:37.000Z",
  "title": "Generative Modeling of Convolutional Neural Networks",
  "authors": [
    "Jifeng Dai",
    "Ying-Nian Wu"
  ],
  "categories": [
    "cs.CV",
    "cs.LG",
    "cs.NE"
  ],
  "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.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2014-12-19T11:34:37.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "convolutional neural networks",
      "generative model",
      "gradient pre-training consistently helps",
      "visualization method generates meaningful",
      "generative visualization method"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable",
      "adsabs": "2014arXiv1412.6296D"
    }
  }
}