TinyTL: Reduce Memory, Not Parameters
         for Efficient On-Device Learning

              Han Cai1 , Chuang Gan2 , Ligeng Zhu1 , Song Han1
         1
           Massachusetts Institute of Technology, 2 MIT-IBM Watson AI Lab
               {hancai, chuangg, ligeng, songhan}@mit.edu

                        Abstract
      Efficient on-device learning requires a small memory footprint at training time to
      fit the tight memory constraint. Existing work solves this problem by reducing
      ...