Benchmarking Vision on Edge vs Cloud
Data Skeptic
00:00
Approximation Knobs
For se an n kernel that is used as the object detective for approxdet. So if you see there is, one of the knobs that we have shown in this paper is the number of proposals. A high ladency is bad, and a higher curacy is good. But you can't get both at the same time, so you might have to cut down a ladency by also cutting down on the curacy some. And i think this kind of one time adaptation to both content and contention, it's basically the only thing in the market.
Transcript
Play full episode
