Machine Learning for Personalised Health
summary
Probabilistic modelling and latent‑variable graphical models are used to disaggregate heterogeneous disease into actionable subgroups, applied with longitudinal logistic and hidden‑Markov style models and implemented in Infer.NET and compared to tools like Stan and Edward. On a 2,000‑child cohort five asthma subtypes emerge including a 3% persistent troublesome group with poor treatment response, and on 12,000 children only 3% follow the classic allergic march. A wearable prototype with tiny coin motors called EMMA shows observable improvement in a tremor writing demo.
AI-written summary · about
chapters (8)
- 0:00Personalized health with ML
- 2:11Personalized healthcare importance
- 4:42Endotype discovery in diseases
- 9:25Asthma endotypes research
- 15:25Allergic March hypothesis findings
- 20:02ML healthcare challenges
- 22:58Emma project demo for tremors
- 25:08Research impact summary
related talks
20mDiagnostic Medical application of ML
18mMedical AI Beyond Prediction: From Normative Learning to Agentic Healthcare AI
19mArya
39mBest Practices for Prototyping ML Models for Healthcare
33mLessons Learned Applying Large Language Models in Healthcare
39mBridging health inequalities through machine learning
28mAI in healthcare - Deep learning for precision psychiatry
20mMachine Learning in Cardiology