2019 CDM CIO Summit AI Driven Development
0 likes181 views
This document discusses AI-driven development and summarizes key points from a presentation on the topic. The presentation covers: 1) The large market opportunity for AI, with McKinsey estimating $3.5-5.8 trillion potential value; 2) How building ML on existing data strategies requires fetching, cleaning, transforming data to train models, evaluate them, and deploy in production while monitoring; 3) Ways to accelerate business ROI including using hybrid cloud, addressing technical debt beyond just ML code, and ensuring the right skillsets are available.
2019 CDM CIO Summit AI Driven Development
- 1. AI-Driven Development (AutoML) Mar 26th 2019 ? Deep Learning Machine Learning Artificial Intelligence What’s next? GundlapalliC Chandra-Gundlapalli
- 2. Agenda (help accelerate Think Tank discussion) AI Data Strategy Big Picture#01 Time-to- Market Levers #02 Lessons Learned Recap #03 AI = Prediction + Automation + Optimization 2
- 3. #01 Quick look at AI market opportunity big picture 3 Key takeaway # McKinsey: AI $3.5 - $5.8 trillion potential, or 40 percent from all analytics # >30% revenue increase for the AI front runners Proven AI in the market today Sales Retail Finance Health • Omnichannel • Recommendation • Prescriptive sales • Robo-Advisory • Fraud Detection • Billing Exceptions • Call routing • Voice auth • Social listening • Patient analytics • Imaging insights • Drug effectiveness Heliograf Procurement Smart Grid Investor services Zenbo nanny ROI-driven use cases
- 4. #Building ML on top of existing data tech strategy Fetch data Clean data Transform data Train model Evaluate model Prod Deploy Monitor Machine Learning BuildDeploy Train Batch What happened? Real-time What is happening now? Inferences What should we do? Key takeaway Expected Reality(2-3Xexpectedduration) Data Lake, Hub Analytics 4
- 5. #03 ML DNA – accelerate business ROI Services Frameworks Platforms Infrastructure Key takeaway # Build hybrid cloud with scalability & “Bring Your Own” flexibility in each of the layers. # Address hidden tech debt (ML code is only a small fraction) Vision Video Natural Lang Translate Conv BOT ??? AWS SageMaker Google AutoML Kubefow AI-ble Azure ML ??? MXNet TensorFlow PyTorch R Spark ??? GPUTPU Containers (DockerK ubernetes) In-memory Xilink FPGAs Serverless ??? 5 Skillset Citizen data scientists Top coder Rent a data scientist Domain SME DataOps engineers ???
- 6. #Lessons Learned #2 Collaboration 5. Embrace DataOps clarity of purpose (not DevOps) 6. Bias - Business Impact (ROI) vs. Model Accuracy 7. Unified data – discovery & Access unaltered raw data. 8. Data Governance (quality) automation #1 Talent 1. Empower (Value mindset) Data Scientists 2. Talent Gap & Prototyping (Top coder) - status quo 3. Trust & Transparency - Fear & Misunderstanding 4. TEAM RACI – SMEScientistEngineers #3 Deployment 9. Open Data Format & data virtualization latency 10.on-premise multitenancy for future hybrid cloud 11. Train and Run anywhere (cloudedge) – AWS Neo 12. Model Explainability & PROD-like data copies 6
- 7. #02 ML Time-to-Market levers Speed Quality Cost #1 Zero setup • Serverless plug and play. • Open source pre-installed. • Data governance pre-built. #2 Optimized Algorithms • Streaming datasets. • Bring Your Own Algorithm. • Very large datasets. #3 Easier Training • Hyper parameterization • Bring Your Own Container. • Bring Your Own Script. #4 Operationalization • One step deploy • AB testing. • Bring Your Own Model. Key takeaway # BUY & BUILD Solution exists in market today cutting down efforts by >50% # Narrow gap between 1 billion workers & 1 million data experts 7
- 8. #03 AWS SageMaker – “Customer Next Best Action” use case demo 8
- 9. Closing The Loop… 9 Deep Learning Machine Learning Artificial Intelligence Neural nets to design neural nets
Editor's Notes
- #2: AI: The theory & development of computer systems that able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making and translation between languages. ML: Machine learning is a subset of artificial intelligence in the field of computer science that often uses statistical techniques to give computers the ability to “learn” (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed. DL: Deep learning is a subset of machine learning in Artificial Intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured or unlabeled. Also known as Deep Neural Learning or Deep Neural Network.