{
  "id": "1606.06959",
  "version": "v1",
  "published": "2016-06-22T14:10:47.000Z",
  "updated": "2016-06-22T14:10:47.000Z",
  "title": "Dealing with a large number of classes -- Likelihood, Discrimination or Ranking?",
  "authors": [
    "David Barber",
    "Aleksander Botev"
  ],
  "categories": [
    "stat.ML"
  ],
  "abstract": "We consider training probabilistic classifiers in the case of a large number of classes. The number of classes is assumed too large to perform exact normalisation over all classes. To account for this we consider a simple approach that directly approximates the likelihood. We show that this simple approach works well on toy problems and is competitive with recently introduced alternative non-likelihood based approximations. Furthermore, we relate this approach to a simple ranking objective. This leads us to suggest a specific setting for the optimal threshold in the ranking objective.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2016-06-22T14:10:47.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "large number",
      "likelihood",
      "discrimination",
      "perform exact normalisation",
      "simple approach works"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}