DNA-TEQ: An Adaptive Exponential Quantization of Tensors for DNN Inference

TL;DR

DNA-TEQ introduces an adaptive exponential quantization method for DNN tensors, significantly reducing bit-width and energy consumption while maintaining accuracy, enabling efficient deployment on embedded systems.

Contribution

The paper proposes DNA-TEQ, a novel exponential quantization scheme that adaptively quantizes tensors based on their distribution, outperforming linear methods in compression and energy efficiency.

Findings

Achieves 40% average compression over INT8 baseline.

Reduces energy consumption by 66% in dot-product operations.

Maintains high accuracy without retraining DNNs.

Abstract

Quantization is commonly used in Deep Neural Networks (DNNs) to reduce the storage and computational complexity by decreasing the arithmetical precision of activations and weights, a.k.a. tensors. Efficient hardware architectures employ linear quantization to enable the deployment of recent DNNs onto embedded systems and mobile devices. However, linear uniform quantization cannot usually reduce the numerical precision to less than 8 bits without sacrificing high performance in terms of model accuracy. The performance loss is due to the fact that tensors do not follow uniform distributions. In this paper, we show that a significant amount of tensors fit into an exponential distribution. Then, we propose DNA-TEQ to exponentially quantize DNN tensors with an adaptive scheme that achieves the best trade-off between numerical precision and accuracy loss. The experimental results show that DNA-TEQ provides a much lower quantization bit-width compared to previous proposals, resulting in an average compression ratio of 40% over the linear INT8 baseline, with negligible accuracy loss and without retraining the DNNs. Besides, DNA-TEQ leads the way in performing dot-product operations in the exponential domain, which saves 66% of energy consumption on average for a set of widely used DNNs.