Paul M. HeiderKeynote:
Why epidemics are harder to forecast than the weather? (Jeff Shaman has a good graphic)
- uncertainty only exists in the future for weather
- uncertainty is in both the past *and* futre for epidemiology
Keynote:
Common Approaches to Epidemic Forecasting:
- Mechanistic Approaches (you're painting a picture, telling a story)
-- compartmental models (e.g., S-I-R, S-E-I-R-S)
-- agent-based models (individual-level simulation)
- Non-Mechanistic (no story)
-- statistical/machine learning, e.g., SARIMA
- Semi-Mechanistic: data assimilation
- Wisdowm-of-the-Crowds: aggregated human judgement
Keynote:
French group in the 2000's was first to post flu forecasts in real time. (Commentary: wow, that seems so recent.)
Keynote:
Weather forecasting took about 160 years to get as "good" as we are at it now.
Keynote:
What kind of epidemics are forecastable?
- one of a kind events? No. Too rare to meaningfully evaluate.
- Across seasons? Eventually but not the focus of speakers work.
- Within season or short term? Yes.
Keynote:
Why don't we have local health/epidemic forecasts exactly like local weather reports?
Keynote Intro:
AI Themes in submission - implementation, evaluation, maturity, ethics
#AMIAIS25 Keynote Intro: "Given the current policy shifts, we're being disrupted."