The work “Deep Neural Network Benchmarks for Selective Classification” has been accepted for pubblication in The Journal of Data-Centric Machine-Learning Research (DMLR).
This work provides the first benchmark for selective classification with deep neural networks. We provide a comprehensive evaluation of the most popular deep learning architectures according to different criteria. The code can be found here.
A final thanks to my coauthors Lorenzo Perini, Jesse Davis and Salvatore Ruggieri for their help.