Getting to timely insights - how to make it happen?
Download as PPTX, PDF0 likes79 views
The document discusses the challenges and opportunities organizations face in leveraging machine learning to remain competitive amid disruption from digital-native companies and evolving business models. It highlights examples of machine learning applications across various industries, including finance, retail, and utilities, showcasing how these solutions can improve decision-making and operational efficiency. Additionally, it emphasizes the importance of having the right infrastructure and data management strategies to derive timely insights from expansive datasets.
Getting to timely insights - how to make it happen?
- 1. Getting to timely insights How to make it happen? Mandie Quartly, PhD WW Lead, Machine Learning & High Performance Analytics OpenPOWER ISV Ecosystem, IBM @mandieq June 2017
- 2. Are you a disruptor? Or about to be disrupted? 2
- 4. Estimated worldwide startups each day 274,000 We are all vulnerable to seismic shifts External Threats Born-on-digital companies that steal market share or rewrite customer expectations New business models that reinvent our industry and change the game altogether Internal Threats Siloed data and systems Gaps in expertise and skills Inability to react quickly 4
- 5. Why are we here?
- 6. Why are you here? We are actively using Machine Learning in production – living the dream We are testing Machine Learning – how could we benefit? We are investigating using Machine Learning – what’s all the hype about? What’s Machine Learning? - here for the beer
- 7. “AI will save us all”
- 8. 8 When sci-fi becomes reality https://arxiv.org/pdf/1612.03242.pdf https://arxiv.org/pdf/1702.00783.pdf Creating images from textual descriptions Enhance images from low-res versions
- 10. June 2017 Who is doing it?
- 11. A bank in Asia 11 Business objective: Use existing enterprise data to make better predictions: • Which customers to offer larger credit limits for consumer banking • Opportunities to cross sell, up sell Challenge: Transform data from multiple sources to something consumable for training Solution: • Data ETL / Curation / Feature Engineering using Spark • Fed into own DL models created used Tensorflow as part of PowerAI
- 12. A financial institution in China Business objective: • Unattended ATMs become target of increasing crimes • Face occlusion detection could help recognize potential criminal actions and then trigger alarm or limit function • Traditional pattern identification algorithms are not so effective to resolve such many diversified and changing possibilities Solution: Real time video recording and auto-detect face occlusion, support multiple ways of recognition simultaneously. Based upon Caffe as part of PowerAI. (red frame means a face occlusion)
- 13. Outdoor retailer in North America 13 Business objective: • Online shoppers can often abandon their carts due to being overwhelmed by too many options. • In a brick-and-mortar store, consumers can ask for advice when struggling with indecision, whereas a website provides little to no personal assistance. Solution: • A cognitive discovery engine that curates content for each shopper. • Helps shoppers find the right clothing, asking questions about the planned activity, season, weather conditions and other preferences. • Based upon solution developed using IBM Watson Knowledge Studio.
- 14. Utility company in Korea 14 Business objective: • Around 40,000 high voltage power transmission towers requiring visual inspection, often situated in remote or difficult terrain. • Looking to reduce risk of accidents and raise inspection rates. Solution: • Prototype for training a computer vision model for infrastructure inspection. • When developed, the inference model will in real time analyze video data from drones flying over infrastructure to identify failed or failing components for repair or replacement. • Solution based on PowerAI
- 15. Getting to accurate insights fast The ingredients are key 15
- 16. The right ingredients to get to timely insights What’s the question you are looking to answer? How to make it happen? Dive into the detail or abstract away from it? What have you got? What do you need? Who could you partner with? Where? On prem / cloud / hybrid? 16 Infrastructure Data AI / Cognitive Applications / solutions
- 17. AI / Cognitive capabilities REASON They can reason, grasp underlying concepts, form hypotheses, and infer and extract ideas. UNDERSTAND Cognitive systems understand imagery, language and other unstructured data like humans do. LEARN With each data point, interaction and outcome, they develop and sharpen their expertise, so they never stop learning. INTERACT With abilities to see, talk and hear, cognitive systems interact with humans in a natural way. 17
- 18. Making it so… which route to choose? BUILD API BUY 18
- 19. 44 zettabytes unstructured data structured data 2010 2020 DataGrowth Data holds competitive value You are here Data you possess + Data outside your firewall + Data that is coming
- 21. 21
- 22. Moore’s Law Not Holding Up 2000 2020 Performance/$ The Processor is No Longer the Only Source of IT Innovation
- 23. 2000 2020 Memory Storage I/O Attach Accelerators They’re Closing the Gap Performance/$
- 25. System / Integration I/O / Storage / Acceleration Boards / Systems Chip / SOC Software Implementation / HPC / Research
- 26. System / Integration I/O / Storage / Acceleration Boards / Systems Chip / SOC Software Implementation / HPC / Research 300+Members 31Countries 40+ISVs Active Membership From All Layers of the Stack 100K+ Linux Applications Running on Power 2300 ISVs Written Code on Linux Partners Bring Systems to Market 100+ OpenPOWER Ready Certified Products 20+ Systems Manufacturers 40+ POWER-based systems shipping or in development 100+ Collaborative innovations under way
- 27. Open allows you to create what you need
- 28. Who cares about the platform?? 28 “World’s Fastest AI Platform for Enterprise” GPUs - NVLink FPGAs - CAPI Memory bandwidth & cache POWER9, OpenCAPI & PCI Gen4, NVLink 2.0 Open Ecosystem Innovation POWER8, up to 96 threads Get to more accurate insights faster Get more bang for your buck (and keep doing so)
- 29. PowerAI Deep Learning Software Distribution Caffe NVCaffe TorchIBMCaffe Distributed TensorFlowTensorFlow OpenBLAS Theano Deep Learning Frameworks Bazel DIGITSNCCL Distributed Communications Supporting Libraries Chainer http://ibm.biz/powerai
- 30. If you aren’t disrupting, are you about to be disrupted? Have you got the right ingredients? Is your platform open, innovative and going to get you to where you need, when you need? 31
- 31. Thank you Mandie Quartly, PhD @mandieq mandie_quartly@uk.ibm.com Thank you! Let’s get disrupting…
- 32. © IBM Corporation 2017 IBM, the IBM logo, ibm.com, and Watson are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with the appropriate symbol (® or ™), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at “Copyright and trademark information.” • Other company, product, and service names may be trademarks or service marks of others. • References in this publication to IBM products or services do not imply that IBM intends to make them available in all countries in which IBM operates. Trademarks and notes 33Group Name / DOC ID / Month XX, 2017
- 34. PowerAI Deep Learning Software Stack 35 DL Frameworks (TF, Caffe, etc) Data Prep & ETL via Spectrum Conductor for Spark Input Data Deep Learning GUI Data & Model Management, ETL Tools, Monitor, Visualize, Advise DL Insight Tuning Engine AI Vision Computer Vision App Development Toolkit IBM Spectrum Conductor for Spark System mgmt, Distributed ETL, Distributed Training, Hyper-Parameter Optimization Distributed Training
- 35. Reconfigurable hardware Task customized Low latency & power Workload Accelerators + POWER8 FPGA GPU Uses: • Compression • Encryption • high speed streaming search • Monte Carlo simulations CAPI 1000s of simple cores High bandwidth, floating point, and parallelism NVIDIA NVLink Uses: • Deep Neural Networks • Speech Recognition • Chemistry Simulations • JAVA • Hadoop • Graphics
- 36. POWER, the ONLY CPU with NVLink CPU PCIe Attached CPU to P100 GPU Graphics Memory DDR4 PCIeGen3 32GB/s PCIe Data Pipe GPU P8NVL CPU GPU GPU DDR4 80 GB/s NVLinkGraphics Memory Graphics Memory POWER8 NVLink Data Pipe NVLink Attached CPU to P100 GPU 2.8x Bandwidth
- 37. ASIC & FPGA OpenCAPI 2.0 NVLink 2.0 PCIe Gen4 Network & Storage AttachCompute Accelerators V100 GPU 10x Advantage 2x Advantage Flash NVDimms DDR Network & NICs Power9, the Next Generation CPU Designed for Accelerated Computing P9 CAPI 2.0
- 38. P8/ P9 CAPI over PCIeG3,4 / OpenCAPI CAPI Coherent Accelerator Processor Interface Secure, trusted, and virtualized Greater bandwidth & access to your data Enables applications not possible on I/O Accelerated Functions: Analytics close to the data pool Removes device driver and it’s code stack
- 39. Higher Performance with POWER8 CPU - P100 GPU NVLink PowerAI leverages NVLink between CPUs and GPUs to enable fast memory access to large data sets in system memory Two NVLink connections between each GPU and CPU-GPU leads to faster data exchange Large NN models benefit the most P100 GPU POWER8 CPU GPU Memory System Memory P100 GPU 80 GB/s GPU Memory NVLink 115 GB/s P100 GPU POWER8 CPU GPU Memory System Memory P100 GPU 80 GB/s GPU Memory NVLink 115 GB/s
- 40. PowerAI Provides Latest DL Frameworks No need to compile from open-source • Tested, binary builds of common Deep Learning frameworks for ease of implementation • Simple, complete installation process documented on ibm.biz/powerai • Future focus on optimizing specific packages for POWER: OpenBLAS, NVIDIA Caffe, TensorFlow, and Torch PowerAI OS Ubuntu 16.04 CUDA 8.0 cuDNN 5.1 Built w/ MASS Yes OpenBLAS 0.2.19 Caffe 1.0 rc5 NVIDIA Caffe 0.14.5 + 0.15.14 IBM Caffe 1.0 rc3 Chainer 1.20.1 NVIDIA DIGITS 5 Torch 7 Theano 0.9 TensorFlow 1.0.0+ 0.12 GPU 4 x P100 Base System S822LC/HPC