1 paper
SeokHyun Seo, Jinwoo Hong, JungWoo Chae +2
Utilizing well-trained representations in transfer learning often results in superior performance and faster convergence compared to training from scratch. However, even if such go…