Quantization Aware Training using the LSQ method

Context

Now that Post Training Quantization is implemented, we want to explore Quantization Aware Training.

A straight forward approach for the QAT is to use the LSQ node, that quantizes both the weights and the activation.

Key steps

• test the LSQ operator (cpu)
• test the LSQ operator (cuda)
• implement the quantizers insertion and init routines
• benchmark the LSQ QAT on the MNIST (cpu)
• benchmark the LSQ QAT on the CIFAR-10 (cuda)
• benchmark the LSQ QAT on the ImageNet (cuda)

Results

TODO