Medical AI Beyond Prediction: From Normative Learning to Agentic Healthcare AI
summary
Describes moving from label-heavy classifiers to normative learning that trains on one class (e.g., healthy scans) to flag out-of-distribution cases, shows reconstruction and latent-embedding methods that localize anomalies as well as manual labels, presents hospital-scale synthetic patients and digital dataset twins that preserve privacy, reports building domain-specific foundation embeddings with 700 GPUs, and demonstrates multi-agent systems that reproduce and slightly improve 20 years of cardiac phenotyping research within an afternoon.
AI-written summary · about
chapters (14)
- 0:00Medical AI research intro
- 1:28UK vs Germany healthcare
- 2:31Manual AI to deep learning
- 3:35Labeling rare diseases
- 6:40Normative learning in AI
- 8:48Anomaly detection with networks
- 10:11Normative ranges interpretation
- 11:35Representation spaces navigation
- 12:39Domain-specific embedding models
- 13:45Synthetic data and digital twins
- 14:28Normative AI in agentic models
- 15:53Clinical research acceleration
- 16:35Emergent properties in AI
- 17:17Future of commoditized AI
related talks
21mGen AI in Health: What's New in 2026?
36mResponsible AI
58mHealthcare with AI of the Future
37mAI in medicine: where do we stand?
22mThe Future of AI and Digitally Enabled Health & Medicine: What's Next
28mThe Future of Healthcare – AI’s Promise and the Challenges Ahead
09mHow AI is changing the healthcare system
22mLeveraging AI/ML for Transformative Software Solutions in Healthcare