Quantization:

The method for representing the weights of models into lower precision in order to maximize the usability and lower the resources consumption.

Quantize and de-quantize a tensor.

method of mapping the large set to the smaller set of values.

what we can quantize in neural networks.

• The weights,
• Activations ( values which propagate in neural networks) .

Advantages of quantization

Linear mapping:

higher precision range to the lower percision range.

$$ r = s(q-z) $$