Data-Centric Debugging: mitigating model failures via targeted data collection
Deep neural networks can be unreliable in the real world when the training set does not adequately cover all the settings where they are deployed. Focusing on image classification, we consider the setting where we have an error distribution $\mathcal{E}$ representing a deployment scenario where the...
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Articolo |
| Pubblicazione: |
2022
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2211.09859 http://arxiv.org/abs/2211.09859 |
| Tags: |
Nessun Tag, puoi essere il primo ad aggiungerne!!
|