Evidential Deep Neural Networks for Uncertain Data Classification - Archive ouverte HAL Accéder directement au contenu
Chapitre D'ouvrage Année : 2020

Evidential Deep Neural Networks for Uncertain Data Classification

Résumé

Uncertain data classification makes it possible to reduce the decision risk through abstaining from classifying uncertain cases. Incorporating this idea into the process of computer aided diagnosis can greatly reduce the risk of misdiagnosis. However, for deep neural networks, most existing models lack a strategy to handle uncertain data and thus suffer the costs of serious classification errors. To tackle this problem, we utilize Dempster-Shafer evidence theory to measure the uncertainty of the prediction output by deep neural networks and thereby propose an uncertain data classification method with evidential deep neural networks (EviNet-UC). The proposed method can effectively improve the recall rate of the risky class through involving the evidence adjustment in the learning objective. Experiments on medical images show that the proposed method is effective to identify uncertain data instances and reduce the decision risk.
Fichier principal
Vignette du fichier
P157.pdf (958.67 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02921338 , version 1 (25-08-2020)

Identifiants

Citer

Bin Yuan, Xiaodong Yue, Ying Lv, Thierry Denoeux. Evidential Deep Neural Networks for Uncertain Data Classification. Knowledge Science, Engineering and Management (Proceedings of KSEM 2020), Springer Verlag, pp.427-437, 2020, Lecture Notes in Computer Science, ⟨10.1007/978-3-030-55393-7_38⟩. ⟨hal-02921338⟩
218 Consultations
285 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More