Quantization-Aware Interval Bound Propagation for Training Certifiably Robust Quantized Neural Networks

We study the problem of training and certifying adversarially robust quantized neural networks (QNNs). Quantization is a technique for making more efficient by running them using low-bit integer arithmetic therefore commonly adopted in industry. Recent work has shown that floating-point have been verified to be can become vulnerable adversarial attacks after quantization, certification representation necessary guarantee robustness. In this work, we present quantization-aware interval bound propagation (QA-IBP), novel method QNNs. Inspired advances learning non-quantized networks, our algorithm computes gradient an abstract actual network. Unlike existing approaches, handle discrete semantics Based on QA-IBP, also develop complete verification procedure verifying robustness QNNs, which guaranteed terminate produce correct answer. Compared key advantage it runs entirely GPU or other accelerator devices. demonstrate experimentally approach significantly outperforms methods establish new state-of-the-art