AMC: AutoML for Model Compression and Acceleration on Mobile Devices

Reference1

AMC: leverage reinforcement learning to efficiently sample the design space.

4x FLOPs reduction, 2.7% better accuracy than hand-crafted model compression for VGG-16 on ImageNet.

speedup 1.53x on GPU (Titan Xp) and 1.95x on Android phone (Google Pixel I), with negligible loss of accuracy.

  1. He, Yihui, Ji Lin, Zhijian Liu, Hanrui Wang, Li-Jia Li, and Song Han. “Amc: Automl for model compression and acceleration on mobile devices.” In Proceedings of the European Conference on Computer Vision (ECCV), pp. 784-800. 2018. ↩
Created May 24, 2020 // Last Updated Aug 31, 2021

If you could revise
the fundmental principles of
computer system design
to improve security...

... what would you change?