.png){kind=icon}
.png){kind=icon}
Return
Self-ensemble for test time adaptation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131038.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We show that for the TTA problem, compared with using a single model, an ensemble of multiple smaller models with fewer or comparable overall parameters can also markedly improve TTA performance. • We propose a novel self-ensemble method, in which ensembling is performed on multiple internal classifiers of a single model. The proposed method only incurs slight increase in parameters, and can be readily integrated into existing TTA methods in a plug-in manner to enhance cross-domain robustness. • Evaluation on benchmark datasets demonstrates that the self-ensemble method can significantly improve the performance of existing state-of-the-art TTA methods. Further, when combining it with the ensembling of smaller models, even greater improvement can be achieved.
Keywords:
TTA
ensemble methods
self-ensemble
cross-domain robustness
model performance