Can I read Efficient Integer-Arithmetic Inference on EtoBox?

Efficient Integer-Arithmetic Inference by Guilherme is a document available to read on EtoBox.

What is Efficient Integer-Arithmetic Inference about?

This document proposes a quantization scheme for neural networks that allows for efficient integer-only arithmetic inference. It quantizes weights and activations to lower bit depths, such as 8-bit integers, and modifies the training procedure to preserve accuracy after quantization. Evaluating this approach on MobileNets, an already efficient model, it significantly improves the tradeoff between accuracy and latency for image classification and object detection on mobile CPUs compared to baseline and prior

Author

Guilherme

Language

EN