Counter-Example Guided Neural Network Quantization Refinement (CEG4N)

Neural networks are essential components of learning-based
software systems. However, their high compute, memory, and power re-
quirements make using them in low resources domains challenging. For
this reason, neural networks are often quantized before deployment. Ex-
isting quantization techniques tend to degrade the network accuracy.
We propose Counter-Example Guided Neural Network Quantization Re-
finement (CEG4N). This technique combines search-based quantization
and equivalence verification: the former minimizes the computational re-
quirements, while the latter guarantees that the network’s output does
not change after quantization. We evaluate CEG4N on a diverse set of
benchmarks that include large and small networks. Our technique was
successful at quantizing the networks in our evaluation while producing
models with up to 72% better accuracy than state-of-the-art techniques.