CAPE: Generalized Convergence Prediction Across Architectures Without Full Training

Publication
Transactions on Machine Learning Research (TMLR)