Position-based Scaled Gradient for Model Quantization and Pruning
This work addresses model compression for efficient deployment in resource-constrained environments, presenting an incremental improvement through a novel gradient scaling technique.
The authors tackled the problem of making neural network weights more amenable to compression by proposing a position-based scaled gradient (PSG) that acts as a regularizer, reducing the gap between full-precision and compressed models, with experimental results on CIFAR-10/100 and ImageNet showing effectiveness in pruning and quantization even at extremely low bits.
We propose the position-based scaled gradient (PSG) that scales the gradient depending on the position of a weight vector to make it more compression-friendly. First, we theoretically show that applying PSG to the standard gradient descent (GD), which is called PSGD, is equivalent to the GD in the warped weight space, a space made by warping the original weight space via an appropriately designed invertible function. Second, we empirically show that PSG acting as a regularizer to a weight vector is favorable for model compression domains such as quantization and pruning. PSG reduces the gap between the weight distributions of a full-precision model and its compressed counterpart. This enables the versatile deployment of a model either as an uncompressed mode or as a compressed mode depending on the availability of resources. The experimental results on CIFAR-10/100 and ImageNet datasets show the effectiveness of the proposed PSG in both domains of pruning and quantization even for extremely low bits. The code is released in Github.