Trustworthy AI for the Physical World
summary
Siemens AI Lab frames industrial ML as deployment problems not research puzzles, reporting lab work on robustness, explainability and continuous learning applied to turbines, trains, compressors and shop‑floor vision. Methods include active learning to cut expert annotation, knowledge graphs to inject domain rules and improve explainability, and compact models for low‑memory edge inference. A proposed AI Ops twin pairs virtual testing with on‑device rollouts for continuous monitoring and domain‑generalization across sites.
AI-written summary · about
chapters (27)
- 0:00Intro
- 1:14Industrial AI
- 2:12Selfish Systems
- 3:02Connected Physical World
- 3:36DataDriven Society
- 4:31Ecosystem Beats
- 5:42Business Impact
- 7:12Data Readiness
- 7:52Challenges
- 9:15Trust
- 10:29Challenges in both worlds
- 11:13Machine learning is a fragile beast
- 11:45Correlation is not causation
- 12:01French is crazy
- 12:32For business
- 13:17Trustworthy predictions
- 13:56Robustness
- 14:53Active Learning
- 15:38Knowledge Graphs
- 16:24Shipping
- 16:36In operation
- 17:15New horizon
- 18:03Feedback
- 19:53Data
- 20:26Risk mitigation
- 21:45Key takeaways
- 23:04Bias explainability
related talks
20mDeploying AI in Days, Not Months: Siemens Industrial Edge on
01mIndustrial AI: Physical & Agentic AI
46mCan you trust your AI?
45mWhat Security Measures Are Necessary to Trust in AI?
24mMachine Intelligence gets real
15mAI Gardening: Harvesting and Pruning the AI Tree
27mAI at Scale in the Enterprise
01h 30mTrustworthy AI & Open Source