{
  "id": "1611.09428",
  "version": "v1",
  "published": "2016-11-28T23:35:25.000Z",
  "updated": "2016-11-28T23:35:25.000Z",
  "title": "Stochastic Thermodynamics of Learning",
  "authors": [
    "Sebastian Goldt",
    "Udo Seifert"
  ],
  "comment": "5 pages, 3 figures, 7 pages of supplemental material",
  "categories": [
    "cond-mat.stat-mech",
    "cond-mat.dis-nn",
    "physics.bio-ph"
  ],
  "abstract": "Virtually every organism gathers information about its noisy environment and builds models from that data, mostly using neural networks. Here, we use stochastic thermodynamics to analyse the learning of a classification rule by a neural network. We show that the information acquired by the network is bounded by the thermodynamic cost of learning and introduce a learning efficiency $\\eta\\le1$. We discuss the conditions for optimal learning and analyse Hebbian learning in the thermodynamic limit.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2016-11-28T23:35:25.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "stochastic thermodynamics",
      "neural network",
      "organism gathers information",
      "thermodynamic cost",
      "noisy environment"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 5,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}