efficient learning
- Blog: towards data science
- Efficient Inference in Deep Learning - Where is the Problem?
(2020, by Amnon Geifman) - Petaflops: 1015 floating point operations per seconds
- CPU: ∼0.1⋅1012
GPU: ∼(1−10)⋅1012 - day ˆ= 86.4⋅103sec
- Moore's law:
performance of hardware doubles every 1.8-2 years
ML requirements: doubling every 3-4 months
