Data science vs real world: friends or foes
summary
Built a Bayesian network to estimate evolving bodily‑injury reserves and a Flask UI, then add an OCR + document classifier (CNN with word embeddings) and entity extraction via Watson Knowledge Studio because claim handlers don’t populate structured fields. For banking, replace many product propensity models with a single next‑best‑action Markov decision / batch reinforcement‑learning approach that optimizes 3‑year customer lifetime value (Q‑function learned by supervised models). The RL solution beats separate propensities on revenue in live campaigns (three pilot campaigns show positive ROI).
AI-written summary · about
chapters (14)
- 0:00Introduction
- 2:20Framework
- 9:45Two examples
- 14:22Deployment
- 15:30Approach
- 18:40Banking example
- 19:40Single model
- 20:25Example
- 21:27Markov decision processes
- 22:46Simplified description
- 24:28Markov decision process
- 25:10Reinforcement learning
- 26:28Results
- 27:05Questions
related talks
35mBank evolution to data science-driven business model
38mReal Life Customer Stories with AI
47mAdvanced Reinforcement Learning & Data Science for Social Impact
18mTransforming Financial Decision Intelligence Through Advanced NNs
46mThe AI dominated world – skills required for successfully navigating it
23mCustom-made games with machine learning and big data
41mSpotlight on AI use cases and innovators around the world
18mKeynote: AI for Financial Services