2309 - 通过 Open Horizo​​n 和 EdgeLake 的联邦学习解锁可扩展的边缘 AI.pdf

1、Orlando,FLOctober 69IBM TechXchange2025Roy Shadmon,AnylogTroy Fine,IBMSanjeev Gupta,IBMIBM TechXchange 2025 conference 2309-Unlock Scalable Edge AI through Federated Learning with Open Horizon and EdgeLakeAgenda01020304050607Intro to Federated Learning(FL)Current State of FLEdgeLakes FL FrameworkDem

2、o and OverviewScaling,Deploying and Securing FL with Open HorizonDemo-in-a-box-edgelake-flSummary,Next Steps,Q&AIntro to EdgeLakeAICloudAITodayEdgeLakeEdgeQueryQueryAIIntel project https:/openfl.readthedocs.io/en/latest/index.html#What is Federated Learning?Key points:Decentralized:Model training ha

3、ppens on local devices instead of a central serverData Privacy&Security:Raw data remains in-place;only model updates are sharedEfficient&Scalable:No raw data transfer,scales horizontallyContinuous learning:Enables models to adapt to local user behaviorUse Cases:Privacy-preserving(healthcare,finance,

4、IT),Localized learningFederated Learning OverviewAgenda010203040506Intro to Federated Learning(FL)Current State of FLEdgeLakes FL FrameworkDemo and OverviewScaling,Deploying and Securing FL with Open HorizonSummary,Next Steps,Q&ACurrent Limitations of FL?Decentralized:No data services at the edgeDat

5、a heterogeneity:No efficient method to unify distributed data Participants heterogeneity:No efficient method to orchestrate many participantsSkills needed:Must manage the entire tech stack:Networking,Databases,Cryptography,Machine LearningOperational complexity:Distributed infrastructure,distribute

6、modelsAggregatorDeployment of federated learning algorithm to each deviceConduct the federated learning process continuouslyNot scalable to 100,1000,10000 nodesAgenda010203040506Intro to Federated Learning(FL)Current State of FLEdgeLakes FL FrameworkDemo and OverviewScaling,Deploying and Securing FL