Mixed Quantization

 In order to achieve the best results, a smart mixture of both

techniques should be utilized.
[1] Saha, S., 2018. A Comprehensive Guide to Convolutional Neural Networks—the ELI5 way. [online] Medium.
Available at: https://towardsdatascience.com/a-comprehensive-guide-to-convolutional-neural-networks-the-eli5-way-
3bd2b1164a53
[2] W. Shi, J. Cao, Q. Zhang, Y. Li and L. Xu, "Edge Computing: Vision and Challenges," in IEEE Internet of Things
Journal, vol. 3, no. 5, pp. 637-646, Oct. 2016, doi: 10.1109/JIOT.2016.2579198.
[3] Coelho, Claudionor N., et al. arXiv: Ultra Low-latency, Low-area Inference Accelerators using Heterogeneous
Deep Quantization with QKeras and hls4ml. No. arXiv: 2006.10159. 2020.
[4] Z. Li et al., "Laius: An 8-Bit Fixed-Point CNN Hardware Inference Engine," 2017 IEEE International Symposium
on Parallel and Distributed Processing with Applications and 2017 IEEE International Conference on Ubiquitous
Computing and Communications (ISPA/IUCC), 2017, pp. 143-150, doi: 10.1109/ISPA/IUCC.2017.00030.
[5] A. Krizhevsky et al. Cifar-10. 2009.
mixed1 mixed2 mixed3 mixed4 mixed5
Activations 8|4|4|4|2 4|6|4|4|4 6|4|2|2|2 4|4|4|2|2 4|2|2|2|2
Weights 8|6|6|6|4|8 8|6|4|4|4|8 8|4|4|2|2|8 4|4|4|2|2|8 4|2|2|2|2|8
 Depending on the accuracy and resource requirements, ideal
configuration can be selected from the given bubble graph.
 For a more detailed analysis of the effect of each quantization
combination, the table shows the change in both the training and
testing accuracy with respect to saved MAC operations and
memory space.