{
  "id": "2111.11223",
  "version": "v2",
  "published": "2021-11-22T14:09:45.000Z",
  "updated": "2022-03-15T17:25:37.000Z",
  "title": "Transfer Learning with Gaussian Processes for Bayesian Optimization",
  "authors": [
    "Petru Tighineanu",
    "Kathrin Skubch",
    "Paul Baireuther",
    "Attila Reiss",
    "Felix Berkenkamp",
    "Julia Vinogradska"
  ],
  "categories": [
    "stat.ML",
    "cs.AI",
    "cs.LG"
  ],
  "abstract": "Bayesian optimization is a powerful paradigm to optimize black-box functions based on scarce and noisy data. Its data efficiency can be further improved by transfer learning from related tasks. While recent transfer models meta-learn a prior based on large amount of data, in the low-data regime methods that exploit the closed-form posterior of Gaussian processes (GPs) have an advantage. In this setting, several analytically tractable transfer-model posteriors have been proposed, but the relative advantages of these methods are not well understood. In this paper, we provide a unified view on hierarchical GP models for transfer learning, which allows us to analyze the relationship between methods. As part of the analysis, we develop a novel closed-form boosted GP transfer model that fits between existing approaches in terms of complexity. We evaluate the performance of the different approaches in large-scale experiments and highlight strengths and weaknesses of the different transfer-learning methods.",
  "revisions": [
    {
      "version": "v2",
      "updated": "2022-03-15T17:25:37.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "transfer learning",
      "gaussian processes",
      "bayesian optimization",
      "closed-form boosted gp transfer model",
      "novel closed-form boosted gp transfer"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}