My current research is primarily driven by the need to adapt AI on edge devices such as mobile devices, IoT devices and autonomous vehicles. The line of work on efficiency robustness, pioneered by our group in 2019, began with the observation that different inputs may incur varied amount of computation costs on neural networks. We have developed various attacks such as white-box attack(CVPR 2020), black-box attack(ICSE 2022), and feed-forward attack(ASE 2022) on a range of applications such as Neural Machine Translation(FSE 2022), Neural Image Caption Generation(CVPR 2022), Transformer-based Language Models(ACL 2023), and Neural ODEs (ICCV-W 2023).