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


Citation Style Language JSON Export

{
  "publisher": "Zenodo", 
  "DOI": "10.5281/zenodo.5253740", 
  "title": "INPUT COMPLEXITY AND OUT-OF-DISTRIBUTION DETECTION WITH LIKELIHOOD-BASED GENERATIVE MODELS", 
  "issued": {
    "date-parts": [
      [
        2020, 
        4, 
        30
      ]
    ]
  }, 
  "abstract": "<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>", 
  "author": [
    {
      "family": "Serra, Joan"
    }, 
    {
      "family": "\u00c1lvarez, David"
    }, 
    {
      "family": "G\u00f3mez, Vicen\u00e7"
    }, 
    {
      "family": "Slizovskaia , Olga"
    }, 
    {
      "family": "F. N\u00fa\u00f1ez, Jos\u00e9"
    }, 
    {
      "family": "Luque, Jordi"
    }
  ], 
  "id": "5253740", 
  "note": "Accepted for ICLR2020", 
  "event-place": "Addiss Abbeba", 
  "type": "paper-conference", 
  "event": "Ninth International Conference on Learning Representation (ICLR 2020)"
}
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