核心概念
DyCE introduces a dynamic configurable early-exit framework for efficient deep learning model compression and scaling.
統計
DyCE는 ResNet152의 계산 복잡성을 23.5%, ConvNextv2-tiny의 계산 복잡성을 25.9% 감소시킴.
引用
"Dynamic compression methods allocate computational resources based on the complexity of each input sample."
"DyCE simplifies the design process of early-exit-based dynamic compression systems."