Perfect Density Models Cannot Guarantee Anomaly Detection
Thanks to the tractability of their likelihood, several deep generative models show promise for seemingly straightforward but important applications like anomaly detection, uncertainty estimation, and active learning. However, the likelihood values empirically attributed to anomalies conflict with t...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2021-12-01
|
| Series: | Entropy |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1099-4300/23/12/1690 |
| _version_ | 1797504941666861056 |
|---|---|
| author | Charline Le Lan Laurent Dinh |
| author_facet | Charline Le Lan Laurent Dinh |
| author_sort | Charline Le Lan |
| collection | DOAJ |
| description | Thanks to the tractability of their likelihood, several deep generative models show promise for seemingly straightforward but important applications like anomaly detection, uncertainty estimation, and active learning. However, the likelihood values empirically attributed to anomalies conflict with the expectations these proposed applications suggest. In this paper, we take a closer look at the behavior of distribution densities through the lens of reparametrization and show that these quantities carry less meaningful information than previously thought, beyond estimation issues or the curse of dimensionality. We conclude that the use of these likelihoods for anomaly detection relies on strong and implicit hypotheses, and highlight the necessity of explicitly formulating these assumptions for reliable anomaly detection. |
| first_indexed | 2024-03-10T04:11:34Z |
| format | Article |
| id | doaj.art-df4fa65d83ad4075a5a490bf765c6c86 |
| institution | Directory Open Access Journal |
| issn | 1099-4300 |
| language | English |
| last_indexed | 2024-03-10T04:11:34Z |
| publishDate | 2021-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Entropy |
| spelling | doaj.art-df4fa65d83ad4075a5a490bf765c6c862023-11-23T08:11:44ZengMDPI AGEntropy1099-43002021-12-012312169010.3390/e23121690Perfect Density Models Cannot Guarantee Anomaly DetectionCharline Le Lan0Laurent Dinh1Department of Statistics, University of Oxford, Oxford OX1 3LB, UKGoogle Research, Montreal, QC H3B 2Y5, CanadaThanks to the tractability of their likelihood, several deep generative models show promise for seemingly straightforward but important applications like anomaly detection, uncertainty estimation, and active learning. However, the likelihood values empirically attributed to anomalies conflict with the expectations these proposed applications suggest. In this paper, we take a closer look at the behavior of distribution densities through the lens of reparametrization and show that these quantities carry less meaningful information than previously thought, beyond estimation issues or the curse of dimensionality. We conclude that the use of these likelihoods for anomaly detection relies on strong and implicit hypotheses, and highlight the necessity of explicitly formulating these assumptions for reliable anomaly detection.https://www.mdpi.com/1099-4300/23/12/1690deep generative modelingprobabilistic modelinganomaly detection |
| spellingShingle | Charline Le Lan Laurent Dinh Perfect Density Models Cannot Guarantee Anomaly Detection Entropy deep generative modeling probabilistic modeling anomaly detection |
| title | Perfect Density Models Cannot Guarantee Anomaly Detection |
| title_full | Perfect Density Models Cannot Guarantee Anomaly Detection |
| title_fullStr | Perfect Density Models Cannot Guarantee Anomaly Detection |
| title_full_unstemmed | Perfect Density Models Cannot Guarantee Anomaly Detection |
| title_short | Perfect Density Models Cannot Guarantee Anomaly Detection |
| title_sort | perfect density models cannot guarantee anomaly detection |
| topic | deep generative modeling probabilistic modeling anomaly detection |
| url | https://www.mdpi.com/1099-4300/23/12/1690 |
| work_keys_str_mv | AT charlinelelan perfectdensitymodelscannotguaranteeanomalydetection AT laurentdinh perfectdensitymodelscannotguaranteeanomalydetection |