The Mean Trap 2010-2020 Learning Models and the Fancy World
Machine learning is becoming increasingly sophisticated, but much of it assumes trust in an average world. Methods that measure performance or risk by standard deviation or a single metric are rapidly rendered impotent in a domain dominated by gaudy fluctuations.
In financial markets and infectious disease forecasting, models that were effective just before suddenly lose their meaning. This is not a failure of the technology, but the result of a design that ignores the fact that the world is nonlinear.
Designing to tolerate extreme events is the condition for survival in a flashy world.
No comments:
Post a Comment