From perter warden blog
accelerator problem:
Intel’s Movidius Neural Compute Stick
- A key observation is that, in a DNN, output size of each node decreases from front-end to back-end, while the change of computation latency is the opposite. (
company list:
- swimming ai
- mimik
- Raspberry Pi Zero
advantage of deploy ML on embedded device:
- a system with no other processes to interrupt your program can make building a mental model of what’s happening very simple
- the straightforward nature of a processor without branch prediction or instruction pipelining makes manual assem‐ bly optimization a lot easier than on more complex CPUs