绿色AI效率评价模型的构建与应用

马晓亮; 刘英; 赵汝强; 杨邦兴; 高洁; 邓从健

doi:10.11959/j.issn.1000-0801.2024184

您当前的位置：

首页 >

文章列表页 >

绿色AI效率评价模型的构建与应用

研究与开发 | 更新时间：2024-09-23

- 绿色AI效率评价模型的构建与应用
- Construction and application of a quantitative efficiency assessment model for green AI in intelligent customer service systems
- 电信科学 2024年40卷第8期页码：130-137
- 作者机构：
  
  1.西安电子科技大学，陕西西安 710071
  2.中国电信股份有限公司广州分公司，广东广州 510620
  3.马晓亮劳模与工匠人才创新工作室，广东广州 510620
  4.广州云趣信息科技有限公司，广东广州 510665
  5.中数通信息有限公司，广东广州 510630
- 作者简介：
  
  [ "马晓亮（1973- ），男，博士，中国电信股份有限公司广州分公司副总经理、正高级工程师，华南理工大学工商管理学院讲席教授，" ]
  [ "刘英（1979- ），男，现就职于中国电信股份有限公司广州分公司，马晓亮劳模和工匠人才创新工作室领衔人，主要研究方向为人工智能、自然语言处理和数据安全保护等。" ]
  [ "赵汝强（1977- ），男，现就职于广州云趣信息科技有限公司，马晓亮劳模和工匠人才创新工作室成员，主要研究方向为智能服务体系搭建和应用、客服中心的系统建设和运营管理、客户投诉和体验管理等。" ]
  [ "杨邦兴（1977- ），男，现就职于中数通信息有限公司，马晓亮劳模和工匠人才创新工作室成员，主要研究方向为智能客服平台组件、语义识别转型。" ]
  [ "高洁（1985- ），女，现就职于中国电信股份有限公司广州分公司，马晓亮劳模和工匠人才创新工作室成员，主要研究方向为机器学习、算法运用等。" ]
  [ "邓从健（1975- ），男，现就职于广州云趣信息科技有限公司，马晓亮劳模和工匠人才创新工作室成员，主要研究方向为自然语言处理、人机交互、算法开发、模式识别等领域。" ]
- 基金信息：
- DOI：10.11959/j.issn.1000-0801.2024184
  中图分类号： TP393
- 收稿日期：2024-04-29，
  
  修回日期：2024-06-12，
  
  纸质出版日期：2024-08-20
- 稿件说明：
移动端阅览
马晓亮,刘英,赵汝强等.绿色AI效率评价模型的构建与应用[J].电信科学,2024,40(08):130-137.

MA Xiaoliang,LIU Ying,ZHAO Ruqiang,et al.Construction and application of a quantitative efficiency assessment model for green AI in intelligent customer service systems[J].Telecommunications Science,2024,40(08):130-137.
马晓亮,刘英,赵汝强等.绿色AI效率评价模型的构建与应用[J].电信科学,2024,40(08):130-137. DOI： 10.11959/j.issn.1000-0801.2024184.

MA Xiaoliang,LIU Ying,ZHAO Ruqiang,et al.Construction and application of a quantitative efficiency assessment model for green AI in intelligent customer service systems[J].Telecommunications Science,2024,40(08):130-137. DOI： 10.11959/j.issn.1000-0801.2024184.

摘要

随着人工智能（artificial intelligence，AI）的兴起，大模型（large language model，LLM）日益成为知识推介和多轮对话的核心技术。伴随而来，AI大模型在数据处理、模型训练和部署过程中的高能耗问题亟须有效评估，以便在模型优化后进行前后量化对比。提出一种AI大模型能耗的评估方法，旨在量化评估AI模型的服务效率（efficiency，E）。该模型使用训练收敛时间（time，T）、模型参数规模（parameter，P）和浮点运算量（floating-point operations，F）等多维度因素，通过构建能源消耗函数

（

，

）实现量化分析；同时，运用非线性最小二乘法，得出模型参数。该分析方法不仅适用于电信运营商客服AI模型的运行效率分析，也可泛化于其他行业的AI模型能耗评估。

Abstract

The rise of AI and the increasing prominence of LLM are positioned as core technologies for knowledge dissemination and multi-turn dialogues. Alongside their growth

the high energy consumption associated with data processing

model training

and deployment in AI large models is necessitating effective evaluation to facilitate quantitative comparisons before and after model optimization. An assessment method for the energy consumption of AI large models was introduced

aimed at quantitatively evaluating the service efficiency (E) of AI models. This model was incorporated with multiple dimensions such as training convergence time (T)

model parameter size (P)

and floating-point operations (F)

and quantitative analysis was achieved through the construction of an energy consumption function C(T

F). Furthermore

by employing the nonlinear least squares method

model parameters were derived. This analysis method was not only applicable to the operational efficiency analysis of AI models used by telecommunications operators but can also be generalized for energy consumption assessment of AI models across various industries.

关键词

Keywords

references

BOCCARDI F , HEATH R W , LOZANO A , et al . Five disruptive technology directions for 5G [J ] . IEEE Communications Magazine , 2014 , 52 ( 2 ): 74 - 80 .

WU Q . Optimization of AI-driven communication systems for green hospitals in sustainable cities [J ] . Sustainable Cities and Society , 2021 , 72 : 103050 .

马晓亮 , 刘英 , 杜德泉 , 等 . 运营商智能客服的关键技术和发展趋势 [J ] . 电信科学 , 2023 , 39 ( 5 ): 76 - 89 .

MA X L , LIU Y , DU D Q , et al . Key technologies and development trends of intelligent customer service for operators [J ] . Telecommunications Science , 2023 , 39 ( 5 ): 76 - 89 .

STRUBELL E , GANESH A , MCCALLUM A . Energy and policy considerations for deep learning in NLP [EB ] . 2019 : 1906 .02243.

魏泽洋 , 刘毅 , 王春艳 , 等 . 环境计算：概念、发展与挑战 [J ] . 清华大学学报(自然科学版) , 2022 , 62 ( 12 ): 1839 - 1850 .

WEI Z Y , LIU Y , WANG C Y , et al . Environmental computing: concept, evolution, and challenges [J ] . Journal of Tsinghua University(Science and Technology) , 2022 , 62 ( 12 ): 1839 - 1850 .

ALLOUHI A , EL FOUIH Y , KOUSKSOU T , et al . Energy consumption and efficiency in buildings: current status and future trends [J ] . Journal of Cleaner Production , 2015 , 109 : 118 - 130 .

SCHWARTZ R , DODGE J , SMITH N A , et al . Green AI [J ] . Communications of the ACM , 2020 , 63 ( 12 ): 54 - 63 .

SOLOW R M . A contribution to the theory of economic growth The Quarterly Journal of Economics , 1956 , 70 ( 1 ): 65 - 94 .

DENIL M , SHAKIBI B , DINH L , et al . Predicting parameters in deep learning [J ] . Advances in Neural Information Processing Systems , 2013 , 26 .

HINTON G , VINYALS O , DEAN J . Distilling the knowledge in a neural network [EB ] . 2015 : 1503 .02531.

HAN S , POOL J , TRAN J , et al . Learning both weights and connections for efficient neural network [J ] . Advances in Neural Information Processing Systems , 2015 , 28 .

GLOROT X , BENGIO Y . Understanding the difficulty of training deep feedforward neural networks [J ] . Journal of Machine Learning Research , 2010 , 9 : 249 - 256 .

SRIVASTAVA N , HINTON G , KRIZHEVSKY A , et al . Dropout: a simple way to prevent neural networks from overfitting [J ] . Journal of Machine Learning Research , 2014 , 15 : 1929 - 1958 .

JIN H , SONG Q , HU X . Auto-keras: an efficient neural architecture search system [C ] // Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . New York : ACM Press , 2019 : 1946 - 1956 .

REN P Z , XIAO Y , CHANG X J , et al . A survey of deep active learning [J ] . ACM Computing Surveys , 2022 , 54 ( 9 ): 1 - 40 .

VINYALS O , BLUNDELL C , LILLICRAP T , et al . Matching networks for one shot learning [J ] . Advances in Neural Information Processing Systems , 2016 : 3637 - 3645 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

多目标滚动时域优化的数学模型在运营商智能派单中的构建与应用