1、Advancing AI:Sustainability for networksReducing energy consumption of AI in networks a pragmatic approachWhite paperThe rapid,global adoption of AI and,especially,todays large language models(LLMs)is evidenced by the explosive growth in usage and funding it has created.However,AI adoption is also p
2、osing environmental and economic challenges due to soaring energy consumption for training and inference.This paper addresses the need for energy efficiency within the telecommunications and networking sectors,where AI is foundational to 6G and network autonomy.We introduce the Energy-efficient AI f
3、or Networks Guide(EA4NG),a pragmatic,three-step framework that ensures AI for networks minimizes its energy footprint and maximizes its energy handprint.Using systematic optimization techniques like pruning,quantization and specialized hardware as well as mandatory consumption monitoring,telecommuni
4、cation providers and AI and data center operators can achieve sustainable AI for networks.The paper advocates for a multi-pronged strategy encompassing brain-inspired AI paradigms and hardware-software co-creation to achieve ambitious energy reduction goals,securing the future profitability and sust
5、ainability of AI deployments.Anne Lee,Gurudutt Hosangadi,Joachim Wabnig,Marc-Olivier Buob,Mikko Honkala,and Sean Kennedy 2White paperAdvancing AI:Sustainability for networksContentsIntroduction 3Goals 5Tenets 5State of the business 5Model compression 6Hardware architectures 6Software architectural a
6、pproaches 7Efficient training methods 7The codesign of training algorithm,model architecture and hardware 7Lessons learned:Energyefficient AI for networks guide(EA4NG)8Step 1:Is AI needed,and,if so,then which one?8Step 2:AI energy optimization 9Step 3:Measuring and monitoring energy consumption 9Bes
0731-84720580
商务合作：really158d
研搜
微信扫码登录
手机快捷登录
账号登录
