Recommended
Recommended
More Related Content
What's hot
What's hot (20)
Similar to FROM BI TO APPLIED AI
Similar to FROM BI TO APPLIED AI (20)
Recently uploaded
Recently uploaded (20)
FROM BI TO APPLIED AI
- 1. FROM BI TO APPLIED AILior Sidi | Braincast.ai CEO KBC conference | 14 Mar 2019
- 2. 2 Agenda 1. Clean table for BI and AI 2. AI for business 3. AI challenges 4. Business for AI + Applied AI
- 3. About Me • Israeli & Half-Bulgarian • Data driven projects • AI, machine learning and Deep learning • (Adversarial, Autoencoders, Explainability) • Braincast.ai
- 4. 4 AI Hype Alert AI is a victim of its own success Extrapolate Misconceptions (sci-fi movies) Lack of education Unrealistic expectations
- 5. 5 “ Five years ago, all the AI researchers were saying: it's much more powerful than you think. And now they're like: it's not as powerful as you think ” Nicholas Thompson, Wired
- 6. 6 Why you need this talk? Value • Get value for the business Education • Separate hype from reality Call for action • You are a critical part of the revolution
- 7. 7 LET’S GET TO KNOW YOU KAHOOT.IT
- 8. 8 What is Business Intelligence? Goal: Aid companies with decision-making Means: Collecting, reporting and analyzing data The key: Quality of data, questions and action Outcome: Impact on core business operation
- 9. 10 BI Challenges Dashboards are not enough People Shortage of experts Human are a bottleneck Data Big data overload Static snapshot Automation - insights in real time Dynamic & Distributed
- 10. 11 BI <-> AI Aid business decision-making Data analytics Data driven organization Quality decision process Automation Self learning Business Intelligence Artificial Intelligence
- 11. 12 What is Artificial Intelligence? • “The science and engineering of making intelligent machines.” John McCarthy • Emulating human cognition in pursuit of problem solving AI is a goal or quest we striving to achieve Applied AI – practical implementations of AI
- 12. 13 Why Now?
- 13. 14 Why Now? Data Availability, flood of sources and labels
- 15. 16 Why Now? Data Research Availability, flood of sources and labels Growing progress in algorithms and theory
- 16. An Analysis of Deep Neural Network Models for Practical Applications, 2017.
- 17. 18 Why Now? Data Power Research Availability, flood of sources and labels Increasing computational power Growing progress in algorithms and theory
- 19. An Analysis of Deep Neural Network Models for Practical Applications, 2017. Facebook managed to reduce the training time of a ResNet-50 deep learning model on ImageNet from 29 hours to one hour Instead of using batches of 256 images with eight GPUs they use batch sizes of 8,192 images distributed across 256 GPUs. Figures copyright Alfredo Canziani, Adam Paszke, Eugenio Culurciello, 2017.
- 20. 21 Why Now? Data Power Research Application Availability, flood of sources and labels Increasing computational power Increasing support from industries Growing progress in algorithms and theory
- 21. 22 “ Era Of Implementation ” Kai-Fu Lee Language Self driving cars Recommendation Cancer Detection
- 22. 23 Value to your costumers Customer service Service recommendation Clarifying a messy picture Costumer segment Credit / insurance fraud Detection in real time Automated decision making Back-office administrative Financial activities Process Automation Cognitive Insight Cognitive Engagement AI for Business
- 23. 24 AI Business Achievements • Increase efficiency • Reduce expenses • Increase customer satisfaction • Improve existing products and services • Create new business opportunities
- 24. 25 BI <-> AI Aid business decision-making Data analytics Data driven organization Quality decision process Automation Self learning Business Intelligence Automation Decision making Model human Intelligence Artificial Intelligence
- 26. 27 Someday Singularity Speculation Decades Human-level intelligence across the range of cognitive tasks Present Perform single task extremely well Artificial Narrow Intelligence (ANI) Artificial General Intelligence (AGI) Artificial Super Intelligence (ASI) Breadth of Intelligence
- 28. 29 First Wave 1970s-1990s Good reasoning But no ability to generalize Thired Wave 2020s-2030s Second Wave 2000s-present Forth Wave 2030 -> Excellent in learning, reasoning and generalize Good learning But No reason and generalize Able to perform any intellectual task a human can Four Waves Of AI
- 29. HowWhenWhat AI ANI Wave 1 Wave 2 Wave 3 AGI Wave 4 ASI
- 31. 33 Learning ability Symbolic AI humans imparting knowledge descriptive rules Sub symbolic execute a task without being explicitly programmed to do so, performance increases with experience -> Machine learning, Search
- 32. 34 Learning ability Learning Tasks • Classification • Continuous estimation • Clustering • Skill acquisition
- 34. 36 Machine Learning • The biggest impact on the world right now • Enabling computers to learn on their own, Iteratively • Spot patterns that humans might miss or never think of in the first place. • Can achieve performance comparable to that of humans without having to imitate human intelligence processes.
- 35. 37 Machine Learning Depends on: • Large data • Labels • Structured inputs
- 36. 38 HOW WOULD YOU CLASSIFY THIS DATA? Age Income
- 37. 39 HOW WOULD YOU CLASSIFY THIS DATA? Age Income
- 38. 40 HOW WOULD YOU CLASSIFY THIS DATA? Age Income
- 39. 41 HOW WOULD YOU CLASSIFY THIS DATA? Age Income
- 41. HowWhenWhat AI ANI Wave 1 Symbolic AI Logic Wave 2 Sub- symbolic AI Machine learning Search Wave 3 AGI Wave 4 ASI
- 42. 44 BI <-> AI Aid business decision-making Data analytics Data driven organization Quality decision process Automation Self learning ? Business Intelligence Automation Decision making Model human Intelligence Learn from experience Implicit learning Automated Artificial Intelligence
- 44. Food classifier
- 45. 47 “ Telling the future by looking at the past assumes that conditions remain constant. This is like driving a car by looking in the rearview mirror. ” - Herb Brod
- 46. 48 AI Challenges Trouble in paradise Erroneous Overfitting, Imbalance, Concept drift, Dimensionality Black swan, Data integrity Humanity Fairness, Labor, Security & Privacy Impact Business Communication Lack of impact, Late go to market Explainability Awareness Education Reduce data-dependentExperience Tools/Tech
- 47. Explainability is hard By Google brain Good luck explaining this
- 48. 50 Explainability is hard Current solutions Informative features Retrain simpler model Decision boundary visualization
- 49. SethBling, MarI/O: Machine Learning for Video Games; 13 June 2015.
- 50. Ribeiro, M. T.,; Guestrin, C. (2016). “ Why Should I Trust You ?” Explaining the Predictions of Any Classifier.
- 51. Goodman, B., & Flaxman, S. (n.d.). European Union regulations on algorithmic and a “‘ right to explanation .’”
- 52. 54 Explainability is hard Current solutions Informative features Retrain simpler model Decision boundary visualization Needed solution Must be loyal to the original model Exact answer with reasoning Human interactive Symbolic?
- 53. Mittu, R., Sofge, D., & Russell, S. (n.d.). Autonomy and Arti cial Intelligence : A Threat or Savior ? Chapter 4: Human Information Interaction, Artificial Intelligence, and Errors
- 54. 56 Value Management Product How you can help AI
- 55. 57 Focus on Value Delivering extraordinary customer value requires deep understanding of the existing business process Customer centricity - Satisfaction, retention, and interaction You create value. Not only communicate value. You don’t define “value” Your costumer do.
- 56. 58 AI Product Management Mistakes to avoid Too specific product Too many pivots Obsession with metrics and analytics Too Generalize Too many POCs
- 57. 59 Data Applied AI Value Management Product Development How you can help AI
- 58. 60 Talk Data Data > Algorithm Garbage in Garbage out Make sure the data is accurate Ask to see and analyze the data
- 59. 61 Applied AI AI is becoming more accessible to domain experts Data scientists will focus on services and advanced stuff Define and understand the features, training, production Feedback the model – adaptive learning Model and incorporate human knowledge less data-dependent AI technology
- 61. 63 Handling feedback Easy Labeling • Education • Invest in GUI
- 63. 65 Handling feedback Easy Labeling • Education • Invest in GUI Model versioning • Capture previous knowledge • stack model • Combine deductive rules / patterns
- 65. 67 Applied AI Braincast.ai • Capture expert knowledge using sequential patterns • Model free method – no training • Good at dynamic complex problems • Handle Temporal, Sparse and multi-stream data • Put the expert within the decision process Fast development cycle Explainable & Accountable Results
- 66. 68 AI Is Cool AI is challenging Get ready to Applied AI Focus on Data & Value Automate decision Efficiency customer satisfaction Improve existing products and services It’s not magic Many places to fail Lack of impact Connect to the data feedback loop Express your knowledge Costumer centric Create new business opportunities Ask the right questions TAKE AWAYS