{
  "id": "1810.05500",
  "version": "v1",
  "published": "2018-10-12T13:37:43.000Z",
  "updated": "2018-10-12T13:37:43.000Z",
  "title": "Predictive Uncertainty through Quantization",
  "authors": [
    "Bastiaan S. Veeling",
    "Rianne van den Berg",
    "Max Welling"
  ],
  "categories": [
    "cs.LG",
    "stat.ML"
  ],
  "abstract": "High-risk domains require reliable confidence estimates from predictive models. Deep latent variable models provide these, but suffer from the rigid variational distributions used for tractable inference, which err on the side of overconfidence. We propose Stochastic Quantized Activation Distributions (SQUAD), which imposes a flexible yet tractable distribution over discretized latent variables. The proposed method is scalable, self-normalizing and sample efficient. We demonstrate that the model fully utilizes the flexible distribution, learns interesting non-linearities, and provides predictive uncertainty of competitive quality.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2018-10-12T13:37:43.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "predictive uncertainty",
      "quantization",
      "stochastic quantized activation distributions",
      "deep latent variable models",
      "rigid variational distributions"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}