Calling Bullshit 5.4: Overfitting

We examine overfitting, the Achilles heel of machine learning, and describe visually what it means and what to look out for.
April 26, 2017
Course: INFO 198 / BIOL 106B. University of Washington
Instructors: Carl T. Bergstrom and Jevin West
Synopsis: Our world is saturated with bullshit. Learn to detect and defuse it.
The course will be offered as a 1-credit seminar this spring through the Information School at the University of Washington. We aim to expand it to a 3 or 4 credit course for 2017-2018. For those who cannot attend in person, we aim to videotape the lectures this spring and make video clips freely available on the web.
callingbullshit.org
/ callin_bull
callinBS
[e-mail]
Information School ischool.uw.edu/
Department of Biology www.biology.washington.edu/
Video edited by Bum Mook Oh
Music by Chris Zabriskie: Prelude No.7

[@emmarector677]

Another component to overfitting is interaction, where there are multiple aspects that are related in and of themselves. As a result, the model starts "double counting" that aspect. Reminds me of the previous videos on correlation and causality.