The deep learning efficient frontier is the curve that determines what is the best accuracy one can achieve given a specific latency, or vice versa. In other words, it is the optimal latency-accuracy tradeoff of deep neural networks on any hardware including cloud, edge, or mobile, for any given task. With AutoNAC, AI teams working on a specific use case can extend its efficient frontier and generate the best model and architecture that are optimized for the production hardware.
Here is an example of a deep learning efficient frontier for object detection models on NVIDIA Jetson Xavier NX. Explore other efficient frontiers for various computer vision tasks and hardware on Deci’s Model Zoo.