Enabling Uncertainty Estimation in Iterative Neural Networks
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912390425083904 |
|---|---|
| author | Durasov, Nikita Oner, Doruk Donier, Jonathan Le, Hieu Fua, Pascal |
| author_facet | Durasov, Nikita Oner, Doruk Donier, Jonathan Le, Hieu Fua, Pascal |
| contents | Turning pass-through network architectures into iterative ones, which use their own output as input, is a well-known approach for boosting performance. In this paper, we argue that such architectures offer an additional benefit: The convergence rate of their successive outputs is highly correlated with the accuracy of the value to which they converge. Thus, we can use the convergence rate as a useful proxy for uncertainty. This results in an approach to uncertainty estimation that provides state-of-the-art estimates at a much lower computational cost than techniques like Ensembles, and without requiring any modifications to the original iterative model. We demonstrate its practical value by embedding it in two application domains: road detection in aerial images and the estimation of aerodynamic properties of 2D and 3D shapes. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_16732 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Enabling Uncertainty Estimation in Iterative Neural Networks Durasov, Nikita Oner, Doruk Donier, Jonathan Le, Hieu Fua, Pascal Artificial Intelligence Turning pass-through network architectures into iterative ones, which use their own output as input, is a well-known approach for boosting performance. In this paper, we argue that such architectures offer an additional benefit: The convergence rate of their successive outputs is highly correlated with the accuracy of the value to which they converge. Thus, we can use the convergence rate as a useful proxy for uncertainty. This results in an approach to uncertainty estimation that provides state-of-the-art estimates at a much lower computational cost than techniques like Ensembles, and without requiring any modifications to the original iterative model. We demonstrate its practical value by embedding it in two application domains: road detection in aerial images and the estimation of aerodynamic properties of 2D and 3D shapes. |
| title | Enabling Uncertainty Estimation in Iterative Neural Networks |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2403.16732 |