{
  "id": "1910.02760",
  "version": "v1",
  "published": "2019-10-07T12:57:45.000Z",
  "updated": "2019-10-07T12:57:45.000Z",
  "title": "Negative Sampling in Variational Autoencoders",
  "authors": [
    "Adrián Csiszárik",
    "Beatrix Benkő",
    "Dániel Varga"
  ],
  "categories": [
    "cs.LG",
    "stat.ML"
  ],
  "abstract": "We propose negative sampling as an approach to improve the notoriously bad out-of-distribution likelihood estimates of Variational Autoencoder models. Our model pushes latent images of negative samples away from the prior. When the source of negative samples is an auxiliary dataset, such a model can vastly improve on baselines when evaluated on OOD detection tasks. Perhaps more surprisingly, we present a fully unsupervised variant that can also significantly improve detection performance: using the output of the generator as negative samples results in a fully unsupervised model that can be interpreted as adversarially trained.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2019-10-07T12:57:45.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "negative sampling",
      "negative samples",
      "notoriously bad out-of-distribution likelihood estimates",
      "model pushes latent images",
      "variational autoencoder models"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}