Artificial Intelligence As a Service
Download as PPTX, PDF0 likes560 views
The document discusses the importance of artificial intelligence (AI) and machine learning (ML) in data science for answering business questions and driving value. It highlights the growth of AI spending and emphasizes that successful AI adoption requires a focus on business value, effective model deployment, and data quality. Additionally, the document suggests leveraging consulting resources and offers tools and strategies for overcoming common data science challenges.
Artificial Intelligence As a Service
- 1. ARTIFICIAL INTELLIGENCE AS A SERVICE John Liu, Intelluron Corp
- 2. Fill in the blank Data Science is:
- 3. Data Science Answering business questions through data.
- 4. Fill in the blank Machine Learning is:
- 5. Machine Learning Ability to improve performance without explicit instruction.
- 6. Fill in the blank Artificial Intelligence is:
- 7. Artificial Intelligence Ability of a machine to replicate intelligent human behavior.
- 8. Worldwide Spending on Cognitive and Artificial Intelligence Systems Will Grow to $19.1 Billion in 2018 Source: IDC
- 9. 40% of digital transformation initiatives will use AI services. AI spending will grow to $52.2 billion in 2021. Source: IDC
- 10. Corporate Adoption AI/ML Awareness AI/ML Adoption AI/ML Native
- 11. 3 big ideas…
- 12. #1 AI Adoption Should Focus on Business Value
- 13. Why Analytics Projects Fail • Lack of focus on business question/ROI • Failing to define metrics and success criteria • Data strategy: data availability, quality, delivery, lineage, storage, access • Failing to address ERPQ
- 14. Do’s and Don’ts Don’t We need AI to stay competitive (FOMO) Do How can advanced analytics improve operating margins
- 16. Answering Business Questions Descriptive • Business Intelligence • SQL • OLAP • Spreadsheets Prescriptive • Data Science • ML • AI
- 17. #2 Business Value is Only Achieved when Models are in Production
- 18. Common Scenario DS I’ve built a model that can … using XLNet and a 1000-layer ResNet model in tensorflow IT Great, but it takes HOW LONG to process a single document on a 1024 GPU cluster?!
- 20. Data Science Cycle Credit: Aaron Lipeles
- 21. Some DS/ML ERPQ Challenges Data Notebook/Code Model Quality check Access Security Versioning Lineage Storage Versioning Documentation Dependencies Versioning Reproducibility Lineage Documentation
- 22. Data Version Control (DVC) https://dataversioncontrol.com/
- 23. CookieCutter Data Science https://drivendata.github.io/cookiecutter-data-science/
- 24. Pachyderm
- 25. MLflow MLflow Tracking MLflow Projects MLflow Models Logging: • parameters, • code versions • metrics • output files in any environment for later comparison. Packaging reusable data science code (including dependences) that can be chained together into multistep workflows. Standardized packaging of ML models for deployment in diverse distributed platforms.
- 27. Model Deployment: Containers Source: Cisco
- 31. #3 Leverage Consulting Resources to Bootstrap whenever possible
- 32. Deployment • Data strategy • Data science • Model training • Prod Infrastructure • Prod Monitoring • Model drift/retrain • Biz Value review
- 33. AI as a Service • Built-in Data Storage/Versioning/Lineage • Platform for Model Building & Tuning • Production-ready Deployment & Integration • Secure and Private platform • Cloud Service – Pay as you go
- 34. Consultants with Expertise • Data Strategy • Open Source AI/ML tools • Cloud Infrastructure • Serverless Architecture • Security
- 35. Last Thoughts • Training Data vs Training Examples – Transfer Learning – Domain Adaptation – Few Shot Learning • Machine Learning with Small Data – Automating Data Science with AI!
- 36. Thank you.