Embedded and Edge AutoML

Model Quantization

Optimization technique that reduces the numerical precision of a neural network's weights and activations (typically from 32-bit to 8-bit or less) to decrease model size and accelerate inference on constrained hardware.

← Wstecz