Ensembling is a very well-known strategy consisting in fusing several different models to achieve a new model for classification or regression tasks. Ensembling has been proven to provide superior performance in various contexts related to pattern recognition and artificial intelligence. The winners of public challenges in image analysis often adopt solutions based on Ensembling. The idea of Ensembling has also provided suggestions for introducing recent deep learning architectures with multiple layer connections that mimic ensembling approaches. However, the full potential offered by Ensembling is not yet fully exploited. This paper aims to explore possible research directions and define new fusion approaches. Preliminary experimental tests show favorable results with an increment in accuracy regarding the number of operations needed in training and inference.
Exploring ensembling in deep learning
Bruno A;Martinelli M;Moroni D
2022
Abstract
Ensembling is a very well-known strategy consisting in fusing several different models to achieve a new model for classification or regression tasks. Ensembling has been proven to provide superior performance in various contexts related to pattern recognition and artificial intelligence. The winners of public challenges in image analysis often adopt solutions based on Ensembling. The idea of Ensembling has also provided suggestions for introducing recent deep learning architectures with multiple layer connections that mimic ensembling approaches. However, the full potential offered by Ensembling is not yet fully exploited. This paper aims to explore possible research directions and define new fusion approaches. Preliminary experimental tests show favorable results with an increment in accuracy regarding the number of operations needed in training and inference.File | Dimensione | Formato | |
---|---|---|---|
prod_468472-doc_188744.pdf
accesso aperto
Descrizione: Preprint - Exploring Ensembling in Deep Learning
Tipologia:
Versione Editoriale (PDF)
Dimensione
239.88 kB
Formato
Adobe PDF
|
239.88 kB | Adobe PDF | Visualizza/Apri |
prod_468472-doc_191269.pdf
non disponibili
Descrizione: Exploring Ensembling in Deep Learning
Tipologia:
Versione Editoriale (PDF)
Dimensione
343.41 kB
Formato
Adobe PDF
|
343.41 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.