{
  "id": "2012.06969",
  "version": "v2",
  "published": "2020-12-13T05:46:46.000Z",
  "updated": "2020-12-16T02:22:46.000Z",
  "title": "Predicting Generalization in Deep Learning via Local Measures of Distortion",
  "authors": [
    "Abhejit Rajagopal",
    "Vamshi C. Madala",
    "Shivkumar Chandrasekaran",
    "Peder E. Z. Larson"
  ],
  "comment": "Added preprint footnote",
  "categories": [
    "stat.ML",
    "cs.LG"
  ],
  "abstract": "We study generalization in deep learning by appealing to complexity measures originally developed in approximation and information theory. While these concepts are challenged by the high-dimensional and data-defined nature of deep learning, we show that simple vector quantization approaches such as PCA, GMMs, and SVMs capture their spirit when applied layer-wise to deep extracted features giving rise to relatively inexpensive complexity measures that correlate well with generalization performance. We discuss our results in 2020 NeurIPS PGDL challenge.",
  "revisions": [
    {
      "version": "v2",
      "updated": "2020-12-16T02:22:46.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "deep learning",
      "local measures",
      "predicting generalization",
      "simple vector quantization approaches",
      "complexity measures"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}