Efficient Bitwidth Search for Practical Mixed Precision Neural Network
This work addresses the problem of optimizing mixed precision quantization for neural networks, which is crucial for practitioners seeking efficient deployment on generic hardware, though it is incremental in improving existing quantization methods.
The paper tackles the challenge of efficiently finding optimal bitwidths for mixed precision neural networks and performing convolutions with varying precisions, proposing an Efficient Bitwidth Search algorithm and a binary decomposition method that outperform handcrafted uniform bitwidth and other mixed precision techniques on CIFAR10 and ImageNet datasets.
Network quantization has rapidly become one of the most widely used methods to compress and accelerate deep neural networks. Recent efforts propose to quantize weights and activations from different layers with different precision to improve the overall performance. However, it is challenging to find the optimal bitwidth (i.e., precision) for weights and activations of each layer efficiently. Meanwhile, it is yet unclear how to perform convolution for weights and activations of different precision efficiently on generic hardware platforms. To resolve these two issues, in this paper, we first propose an Efficient Bitwidth Search (EBS) algorithm, which reuses the meta weights for different quantization bitwidth and thus the strength for each candidate precision can be optimized directly w.r.t the objective without superfluous copies, reducing both the memory and computational cost significantly. Second, we propose a binary decomposition algorithm that converts weights and activations of different precision into binary matrices to make the mixed precision convolution efficient and practical. Experiment results on CIFAR10 and ImageNet datasets demonstrate our mixed precision QNN outperforms the handcrafted uniform bitwidth counterparts and other mixed precision techniques.