An Analysis of Deep Neural Network Models for Practical Applications
Alfredo Canziani、Eugenio Culurciello Weldon School of Biomedical Engineering Purdue University \AndAdam Paszke Faculty of Mathematics, Informatics and Mechanics University of Warsaw
Abstract
Since the emergence of Deep Neural Networks (DNNs) as a prominent technique in the field of computer vision, the ImageNet classification challenge has played a major role in advancing the state-of-the-art. While accuracy figures have steadily increased, the resource utilisation of winning models has not been properly taken into account. In this work, we present a comprehensive analysis of important metrics in practical applications: accuracy, memory footprint, parameters, operations count, inference time and power consumption. Key findings are: (1) power consumption is independent of batch size and architecture; (2) accuracy and inference time are in a hyperbolic relationship; (3) energy constraint is an upper bound on the maximum achievable accuracy and model complexity; (4) the number of operations is a reliable estimate of the inference time. We believe our analysis provides a compelling set of information that helps design and engineer efficient DNNs.
中文速览
深度神经网络在ImageNet图像分类竞赛中准确率不断提升，但研究者们长期忽视了模型实际部署时的资源消耗问题。本文在一块资源受限的嵌入式GPU板（NVIDIA Jetson TX1）上，对AlexNet、VGG、GoogLeNet、ResNet、Inception等多个主流网络从准确率、内存占用、参数量、运算量、推理时间和功耗六个维度进行了系统性横向评测。结果揭示了几个关键规律：功耗与批量大小和网络架构基本无关，满负载时所有网络都消耗约相同的额外功率；准确率与推理吞吐量呈线性关系，换算成推理时间则变为双曲线关系，意味着想再提升一点准确率需要付出大量额外计算；运算次数可以可靠地预估推理时间；能量预算会直接为可达到的最高准确率设定上限。这项工作的价值在于，它为实际落地场景下如何权衡性能与效率提供了量化依据，打破了"只看准确率"的评测惯例，对设计面向边缘端的高效神经网络具有直接的指导意义。
原文 arXiv:1605.07678;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1605.07678v4