Gradient $\ell_1$ Regularization for Quantization Robustness
This addresses the need for flexible and efficient neural network deployment in resource-constrained environments, offering an incremental improvement over existing quantization-aware training methods.
The paper tackles the problem of neural network quantization by proposing a gradient ℓ1 regularization scheme to improve robustness against post-training quantization, enabling a single set of weights to be quantized to different bit-widths on-demand, with experimental validation on CIFAR-10 and ImageNet datasets.
We analyze the effect of quantizing weights and activations of neural networks on their loss and derive a simple regularization scheme that improves robustness against post-training quantization. By training quantization-ready networks, our approach enables storing a single set of weights that can be quantized on-demand to different bit-widths as energy and memory requirements of the application change. Unlike quantization-aware training using the straight-through estimator that only targets a specific bit-width and requires access to training data and pipeline, our regularization-based method paves the way for "on the fly'' post-training quantization to various bit-widths. We show that by modeling quantization as a $\ell_\infty$-bounded perturbation, the first-order term in the loss expansion can be regularized using the $\ell_1$-norm of gradients. We experimentally validate the effectiveness of our regularization scheme on different architectures on CIFAR-10 and ImageNet datasets.