利用 Akida 实现超低 PowerEdge 推理和 OnDevice 学习.pdf

《利用 Akida 实现超低 PowerEdge 推理和 OnDevice 学习.pdf》由会员分享，可在线阅读，更多相关《利用 Akida 实现超低 PowerEdge 推理和 OnDevice 学习.pdf（28页珍藏版）》请在三个皮匠报告上搜索。

1、brainchip Inc Enabling Ultra-Low Power Edge Inference and On-Device Learning with AkidaNandan NayampallyCMOBrainchip Inc.brainchip Inc The Challenge The Approach The Delivery The Results Akida in actionAgenda2brainchip Inc The Challengebrainchip Inc The Problem4$50BAnnual losses in Manufacturing due

2、 tounplanned downtime21TBData generated by 1 Car per day3$6MCosts of training a singleHigh-end model 1$1.1THealthcare cost and lost productivity due to preventable chronic disease44Courtesy:fightchronicdisease.org2Courtesy:D1Courtesy:Spectrum.Ieee.org.“The cost of training,made retraining the model

3、infeasible”3Courtesy:Fbrainchip Inc The Opportunity5Image:Courtesy Pixabay(From Mckinsey AIoT 2030 forecast)$15.7TGlobal GDP Benefit fromAI in 2030*PWC analysis report$1.2TAIoT revenue in 2030*Forbes Business Insights1T+Edge devices in 2030*brainchip Inc The Challenge6Cost of cloud servicesResponsiv

4、eness&reduced latencyScalability&efficiencyPrivacy protection&securitybrainchip Inc The ChallengeThe Solution7Cost of cloud servicesResponsiveness&reduced latencyScalability&efficiencyPrivacy protection&securityReduce cloud inference costMinimize cloud retrainingRapid computation at edgeReal-time co

5、mpute for critical tasksEfficiency within thermal and power budgetsReduced memory and system costPrevent exposure of sensitive dataMinimize raw data being sent to cloudbrainchip Inc The Approachbrainchip Inc Compelling high-performance Extreme efficiencyContinuous learningSecure communication9Event-

6、based processingEvent-based communicationAdvanced spatial-temporal capabilityAt-memory computationEvent-based learningThe Neuromorphic Advantagebrainchip Inc 2ndGenerationFully-digital,neuromorphic,event-based AIUnique ability to learn on device without cloud dependencyWhats New:High performance com