One of the major factors that differentiate Cloud AI from Edge AI is the infrastructure support. Deep neural networks are growing deep and becoming computationally heavy which can be easily supported with cloud infrastructure. But with Edge AI, the neural networks are deployed on resource-constrained devices. To deploy these huge neural architectures on edge devices it is essential to optimize the neural networks.
Optimization is an essential step for Edge AI use case. Optimizing a neural network for an Edge device is a challenging task. It is hardware-specific and requires knowledge about different optimization techniques, hardwares, supported frameworks, and tools. It is also necessary to take care that an acceptable level of accuracy is maintained and there isn’t a significant trade-off between parameters which can negatively affect the performance of the model.