"Better Fine-Tuning by Reducing Representational Collapse."

Armen Aghajanyan et al. (2020)
a service of Schloss Dagstuhl - Leibniz Center for Informatics