In this work, we propose PODNet, approaching incremental learning as rep- resentation learning, with a distillation loss that constrains the evolution of the.
確定! 回上一頁