HEAT: Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection

Abstract

In this work, we study the out-of-distribution (OOD) detection problem through the use of the feature space of a pre-trained deep classifier. We show that learning the density of in-distribution (ID) features with an energy-based models (EBM) leads to competitive detection results. However, we found that the non-mixing of MCMC sampling during the EBM's training undermines its detection performance. To overcome this, we introduce HEAT, an energy-based correction of a mixture of class-conditional Gaussian distributions. We show that HEAT obtains favorable results when compared to a strong baseline like the KNN detector on the CIFAR-10/CIFAR-100 OOD detection benchmarks.

PDF NeurIPSW OpenReview Semantic Scholar

Cite

Text

Lafon et al. "HEAT: Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection." NeurIPS 2022 Workshops: MLSW, 2022.

Markdown

[Lafon et al. "HEAT: Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection." NeurIPS 2022 Workshops: MLSW, 2022.](https://mlanthology.org/neuripsw/2022/lafon2022neuripsw-heat/)

BibTeX

@inproceedings{lafon2022neuripsw-heat,
  title     = {{HEAT: Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection}},
  author    = {Lafon, Marc and Rambour, Clément and Thome, Nicolas},
  booktitle = {NeurIPS 2022 Workshops: MLSW},
  year      = {2022},
  url       = {https://mlanthology.org/neuripsw/2022/lafon2022neuripsw-heat/}
}