人工智能 (AI) 具有渗透和交织在我们社会中每个行业的潜力;然而,即便像 TensorFlow 这样的开源 AI 框架降低了实验新算法的障碍,但如果没有足够的计算能力,所有先进的深度神经网络都将陷入停滞。

Kubeflow 旨在定制 Kubernetes 以支持数据分析工作负载,并为模型开发人员提供弹性 AI 管道,以提高他们的模型开发速度,解决他们的开发过程中的琐事,并构建强大而高效的训綀和服务系统。

我们分享了使用 Kubeflow 为中国国家电网工程师提供培训的过程,该网络是传统企业的典范,可以开发 OCR、愿景和时间序列预测模型,为其财务部门和核心业提供实用的创新功能。

ccone发布于2019/04/19

注脚

展开查看详情

1. Kubeflow & ,

2.Market Competition

3. 2017 $120 2021 $576 Deloitte Global 80 GPU 61% 2019 70% 2030 AI GPU : 230%, 2017 35

4.

5.–

6.–

7. – Exhaust gas temperature of thermal power station Water level of hydropower station

8.– Equipment model Equipment maintenance operation

9.–

10.“ ?” “ ”

11.“ API ”

12.“ ”

13.“400 GPU 20 ”

14.Enterprise Almanac 2018 by Work-Bench

15.Kubeflow Machine Resource Data Collection Data Verification Management Serving Feature ML Infrastructure Code Analysis Tools Extraction Process Configuration Monitoring Management Tools Kubeflow

16.xN Sad example

17. A LOT A LOT

18.• work

19.○ AI API ■ kz docker, kubernetes 1 2 3 4

20.• kz 97% 93% kz 99%

21. Devops (CI/CD) • re-train

22.○ ○ ○ We Need This and This

23. AI OS AI AI SaaS GPU

24. AI OS Streamlined and Data and resource management 1 2 interactive devops pipeline AI OS Integrated “model to business” loop 4 3 Distributed and orchestrated training

25.AI OS

26. AI OS AI AI AI Data Layer

27.AI OS Demo Time

28. AI Solution E-Business Retailing Finance Security Energy Face Object Smart Emotion App Store OCR Recognition Detection Bank Detection Framework Cross-cluster Authentication Backend Kubeflow Admin Storage Manager Data Manager Clever SDK Resource Cross-cluster Intelligent Management Authentication Operation Load Balancer CPU Memory Application Cyclone Automatically GPU Storage Orchestration DevOps Scaling Bare Infra Metal

29.