云计算是为Web开发者构建的,对于深度学习来说,V2是什么样子的?

我们称之为公共云主要是为了管理和部署Web服务器而开发的。这些产品的目标受众是DEV OPS。虽然这是一个庞大而令人兴奋的市场,但数据科学和深度学习的世界非常不同,甚至可能更大。不幸的是,目前可用的工具不是为新的受众设计的,云需要进化。这次演讲将涵盖未来10年的云计算将是什么样子,
展开查看详情

1. 1 / 22 DO NOT DISTRIBUTE Paperspace www.paperspace.com Paperspace Serverless AI for the future of intelligence.

2. 2 / 22 Introduction Deep Learning platform built for developers. Infrastructure automation and software layer to build intelligent applications.

3. 3 / 22 A new generation of AI developers require a rethinking of tooling and workflows.

4. 4 / 22 Why this matters Developers spend 75% of their time managing infrastructure.

5. 5 / 22 So what’s the underlying problem? The cloud was built for a different use- case (web servers) and a different audience (DevOps).

6. 6 / 22 The DevOps ecosystem is rich Traditional Web Services Deep Learning ? ? ? ? ? ? ? + 100s more ? storage, CDN, deploy, monitor, VPC, data, notebooks, train, visualize, load balance, IPsec, CI/CD, DNS ... collaborate, version, hyperparameters ...

7. 7 / 22 The key to solving this problem is finding the right layer of abstraction.

8. 8 / 22 Put Uber/Facebook-grade AI platform in the hands of every developer There is a huge disconnect between modern business objectives and the DL tools that can fulfill them. Business Objective Infrastructure • BI • GPUs • Prediction Closing the Gap • Datastore • Optimization • Algorithms (CNNs, RNNs, ...) • Recommender systems • Frameworks (Pytorch, • Any heuristic TensorFlow, etc)

9. 9 / 22 Paperspace abstracts powerful infrastructure behind a simple software layer making cloud ML as easy as modern web services.

10. 10 / 22 A complete platform for modern deep learning Ingest → Train → Analyze → Deploy + Manage, collaborate, share • Fully-managed GPU infrastructure • Unified dev experience • 1 click Jupyter Notebooks/Lab • API & language integrations • ACL/team controls

11. 11 / 22 GRAI° Model building AI orchestration fabric • Job queuing / management Frameworks Tooling Data • Cloud agnostic • Accelerator architecture native (GPU, FPGA, ASIC, TPU, etc) Keras Pytorch TensorFlow Jupyter Python Quilt • Unified compute • Extensible GRAI° AI orchestration fabric • Built on best practices Job queuing Unified compute Container Data Encryption / (containers, kubernetes, and data deployment management key managment policies) Bare-metal VPS Private cloud (VPC) Public Cloud

12. 12 / 22 Cloudscale with a single line of code Gradient Toolstack > import paperspace as ps # Run job on GPU cluster > ps({‘Type’: ‘TPU’, ‘container’: ‘TensorFlow’ ... }) GRAI Framework Cloud Orchestration & Automation Connecting modern ML and the cloud Cloud Infra by converting infrastructure into code. Raw compute is not sufficient. Network . Storage . Compute

13. 13 / 22 The Pipeline

14. 14 / 22 Remarks from the trenches Trends: 1. Chip renaissance 2. Evolution of ML/AI in practice 3. Consolidation around best practices

15. 15 / 22 Chip renaissance • Graphcore • Cerebras • Nervana • Wave • Google TPU ... The big question today is whether accelerator architectures will follow commodity CPU x86 or lead to a golden era for high-end, use-specific hardware.

16. 16 / 22 The evolution of ML/AI in practice 2016 2018 Consumable API → Refit the Model → Model as core IP • Clarifai • Paperspace • AWS Rekognition • Algorithmia • Google Cloud Vision • FloydHub • MS cognitive services • ClusterOne

17. 17 / 22 Consolidation around best practices • Containerization • Jupyter • Job runner architecture • Pipeline • etc.

18.Paperspace hello@paperspace.com 20 Jay St. Suite 312 www.paperspace.com Brooklyn, NY 11201 (718) 619 4325 Thank you.