Energy awareness in low precision neural networks

TL;DR

This paper introduces PANN, a power-aware neural network approach that accurately models power consumption and enables low-power fixed-precision networks with minimal accuracy loss, adaptable across different power-accuracy trade-offs.

Contribution

The paper develops detailed power consumption models for DNN operations and proposes PANN, a method for converting full-precision networks into low-power variants with flexible power-accuracy trade-offs.

Findings

PANN achieves near full-precision accuracy at low power levels.

The method allows seamless adjustment of power-accuracy trade-offs during deployment.

It outperforms previous quantization methods in power efficiency with minimal accuracy degradation.

Abstract

Power consumption is a major obstacle in the deployment of deep neural networks (DNNs) on end devices. Existing approaches for reducing power consumption rely on quite general principles, including avoidance of multiplication operations and aggressive quantization of weights and activations. However, these methods do not take into account the precise power consumed by each module in the network, and are therefore not optimal. In this paper we develop accurate power consumption models for all arithmetic operations in the DNN, under various working conditions. We reveal several important factors that have been overlooked to date. Based on our analysis, we present PANN (power-aware neural network), a simple approach for approximating any full-precision network by a low-power fixed-precision variant. Our method can be applied to a pre-trained network, and can also be used during training to achieve improved performance. In contrast to previous methods, PANN incurs only a minor degradation in accuracy w.r.t. the full-precision version of the network, even when working at the power-budget of a 2-bit quantized variant. In addition, our scheme enables to seamlessly traverse the power-accuracy trade-off at deployment time, which is a major advantage over existing quantization methods that are constrained to specific bit widths.