Finding Non-Uniform Quantization Schemes Using Multi-Task Gaussian Processes
Abstract
We propose a novel method for neural network quantization that casts the neural architecture search problem as one of hyperparameter search to find non-uniform bit distributions throughout the layers of a CNN. We perform the search assuming a Multi-Task Gaussian Processes prior, which splits the problem to multiple tasks, each corresponding to different number of training epochs, and explore the space by sampling those configurations that yield maximum information. We then show that with significantly lower precision in the last layers we achieve a minimal loss of accuracy with appreciable memory savings. We test our findings on the CIFAR10 and ImageNet datasets using the VGG, ResNet and GoogLeNet architectures.
Cite
Text
do Nascimento et al. "Finding Non-Uniform Quantization Schemes Using Multi-Task Gaussian Processes." Proceedings of the European Conference on Computer Vision (ECCV), 2020. doi:10.1007/978-3-030-58520-4_23Markdown
[do Nascimento et al. "Finding Non-Uniform Quantization Schemes Using Multi-Task Gaussian Processes." Proceedings of the European Conference on Computer Vision (ECCV), 2020.](https://mlanthology.org/eccv/2020/donascimento2020eccv-finding/) doi:10.1007/978-3-030-58520-4_23BibTeX
@inproceedings{donascimento2020eccv-finding,
title = {{Finding Non-Uniform Quantization Schemes Using Multi-Task Gaussian Processes}},
author = {do Nascimento, Marcelo Gennari and Costain, Theo W. and Prisacariu, Victor Adrian},
booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
year = {2020},
doi = {10.1007/978-3-030-58520-4_23},
url = {https://mlanthology.org/eccv/2020/donascimento2020eccv-finding/}
}