Conference paper Open Access

INPUT COMPLEXITY AND OUT-OF-DISTRIBUTION DETECTION WITH LIKELIHOOD-BASED GENERATIVE MODELS

Serra, Joan; Álvarez, David; Gómez, Vicenç; Slizovskaia , Olga; F. Núñez, José; Luque, Jordi


JSON Export

{
  "files": [
    {
      "links": {
        "self": "https://zenodo.org/api/files/f7c9b5f7-9fe4-4fd9-84a7-acf59d79940e/input_complexity_and_out_of_di.pdf"
      }, 
      "checksum": "md5:fa030c9fd1c34044adfdcce3c9cbcdb8", 
      "bucket": "f7c9b5f7-9fe4-4fd9-84a7-acf59d79940e", 
      "key": "input_complexity_and_out_of_di.pdf", 
      "type": "pdf", 
      "size": 1941308
    }
  ], 
  "owners": [
    249507
  ], 
  "doi": "10.5281/zenodo.5253740", 
  "stats": {
    "version_unique_downloads": 12.0, 
    "unique_views": 14.0, 
    "views": 17.0, 
    "version_views": 17.0, 
    "unique_downloads": 12.0, 
    "version_unique_views": 14.0, 
    "volume": 23295696.0, 
    "version_downloads": 12.0, 
    "downloads": 12.0, 
    "version_volume": 23295696.0
  }, 
  "links": {
    "doi": "https://doi.org/10.5281/zenodo.5253740", 
    "conceptdoi": "https://doi.org/10.5281/zenodo.5253739", 
    "bucket": "https://zenodo.org/api/files/f7c9b5f7-9fe4-4fd9-84a7-acf59d79940e", 
    "conceptbadge": "https://zenodo.org/badge/doi/10.5281/zenodo.5253739.svg", 
    "html": "https://zenodo.org/record/5253740", 
    "latest_html": "https://zenodo.org/record/5253740", 
    "badge": "https://zenodo.org/badge/doi/10.5281/zenodo.5253740.svg", 
    "latest": "https://zenodo.org/api/records/5253740"
  }, 
  "conceptdoi": "10.5281/zenodo.5253739", 
  "created": "2021-08-25T06:47:03.820004+00:00", 
  "updated": "2021-08-25T13:48:24.288413+00:00", 
  "conceptrecid": "5253739", 
  "revision": 3, 
  "id": 5253740, 
  "metadata": {
    "access_right_category": "success", 
    "doi": "10.5281/zenodo.5253740", 
    "description": "<p>Likelihood-based generative models are a promising resource to detect out-of- distribution (OOD) inputs which could compromise the&nbsp;robustness or reliability of a machine learning system. However, likelihoods derived from such models have been shown to be problematic for detecting certain types of inputs that sig- nificantly differ from training data. In this paper, we pose that this problem is due to the excessive influence that input complexity has in generative models&rsquo; likelihoods. We report a set of experiments supporting this hypothesis, and use an estimate of input complexity to derive an efficient and parameter-free OOD score, which can be seen as a likelihood-ratio, akin to Bayesian model compari- son. We find such score to perform comparably to, or even better than, existing OOD detection approaches under a wide range of data sets, models, model sizes, and complexity estimates.</p>", 
    "license": {
      "id": "CC-BY-4.0"
    }, 
    "title": "INPUT COMPLEXITY AND OUT-OF-DISTRIBUTION DETECTION WITH LIKELIHOOD-BASED GENERATIVE MODELS", 
    "notes": "Accepted for ICLR2020", 
    "relations": {
      "version": [
        {
          "count": 1, 
          "index": 0, 
          "parent": {
            "pid_type": "recid", 
            "pid_value": "5253739"
          }, 
          "is_last": true, 
          "last_child": {
            "pid_type": "recid", 
            "pid_value": "5253740"
          }
        }
      ]
    }, 
    "communities": [
      {
        "id": "accordion"
      }
    ], 
    "grants": [
      {
        "code": "871793", 
        "links": {
          "self": "https://zenodo.org/api/grants/10.13039/501100000780::871793"
        }, 
        "title": "Adaptive edge/cloud compute and network continuum over a heterogeneous sparse edge infrastructure to support nextgen applications", 
        "acronym": "ACCORDION", 
        "program": "Horizon 2020 Framework Programme - Research and Innovation action", 
        "funder": {
          "doi": "10.13039/501100000780", 
          "acronyms": [], 
          "name": "European Commission", 
          "links": {
            "self": "https://zenodo.org/api/funders/10.13039/501100000780"
          }
        }
      }
    ], 
    "keywords": [], 
    "publication_date": "2020-04-30", 
    "creators": [
      {
        "affiliation": "Dolby Laboratories,  Barcelona, Spain", 
        "name": "Serra, Joan"
      }, 
      {
        "affiliation": "Universitat Polit\u00e8cnica de Catalunya,  Barcelona, Spain", 
        "name": "\u00c1lvarez, David"
      }, 
      {
        "affiliation": "Universitat Pompeu Fabra, Barcelona, Spain", 
        "name": "G\u00f3mez, Vicen\u00e7"
      }, 
      {
        "affiliation": "Universitat Pompeu Fabra, Barcelona, Spain", 
        "name": "Slizovskaia , Olga"
      }, 
      {
        "affiliation": "Universitat Pompeu Fabra, Barcelona, Spain", 
        "name": "F. N\u00fa\u00f1ez, Jos\u00e9"
      }, 
      {
        "affiliation": "Telef\u00f3nica I+D, Research, Barcelona, Spain", 
        "name": "Luque, Jordi"
      }
    ], 
    "meeting": {
      "dates": "May 3rd through the 7th, September, 2021", 
      "title": "Ninth International Conference on Learning Representation", 
      "acronym": "ICLR 2020", 
      "url": "https://iclr.cc", 
      "session": "https://iclr.cc/virtual_2020/poster_SyxIWpVYvr.html", 
      "place": "Addiss Abbeba"
    }, 
    "access_right": "open", 
    "resource_type": {
      "subtype": "conferencepaper", 
      "type": "publication", 
      "title": "Conference paper"
    }, 
    "related_identifiers": [
      {
        "scheme": "doi", 
        "identifier": "10.5281/zenodo.5253739", 
        "relation": "isVersionOf"
      }
    ]
  }
}
17
12
views
downloads
All versions This version
Views 1717
Downloads 1212
Data volume 23.3 MB23.3 MB
Unique views 1414
Unique downloads 1212

Share

Cite as