Insight about signed symmetric and unsigned symmetric quantization

[padeirocarlos]

I wonder know what is the reason behind the use of the unsigned symmetric quantization for both weight e activation?

For activation it can be understandable since the ReLU is well suited for one-tailed distributions, such as ReLU activations.
But what is the explanations for weights? the distributions that are roughly symmetric about zero.
Have you ever tried to change signed symmetric quantization for weight? What was the results? I tried but the results seem not been good!