Algorithms Using Local Graph Features to Predict Epidemics

on YouTube ↗52m

" summary

Claims a nonparametric local estimator that uses repeated K‑neighborhood contact tracing queries (constant K and constant number of queries) to estimate outbreak probability, final size and SIR time curves on sequences of convergent, well‑connected graphs. Tools: coupling SIR with percolation, sprinkling and Menger‑type expansion arguments; implementation is O(1), edge‑differentially private. Simulations on Copenhagen Bluetooth/GPS data show K≈3 recovers outbreak probability; confidence narrows with more queries.

AI-written summary · about

" chapters (21)

1. 0:00Intro
2. 0:36Motivation: Infections Spreading Over a Network
3. 1:27The Epidemic Model
4. 2:34Epidemics in Parallel Worlds
5. 3:55Predicting Epidemics Using Network Models
6. 6:07This Talk in a Nutshell: Local information is enough. You don't need a network model!
7. 8:03Local Estimator: Time Dynamics
8. 9:11Local Estimator: Relative Size of an Outbreak
9. 12:49Epidemics Starting from One Node
10. 14:29Epidemics in Well-Connected Network
11. 15:32Local Estimator: Probability of Outbreak
12. 16:37Relative Size and Probability of Outbreak
13. 20:19Main Technique: Graph Convergence
14. 21:58Large-set Expansion
15. 22:32Examples of Convergent Large-set Expanders
16. 24:58Does the Algorithm Work on Real-world Graphs?
17. 30:03SIR with constant recovery time and percolation
18. 30:44Relative Size of the Giant in Expanders
19. 31:22Proof Sketch: Size of the Giant Converges
20. 33:29Brief History of Sprinkling
21. 35:13Step 2: Sprinkling