Shortcuts: Neural Networks Love to Cheat

Interested in contributing?

Learn more about how you can write and get involved with The Gradient by filling out this short form: Writing + Contributing
Interview about the origins and purpose of The Gradient

Shortcuts: How Neural Networks Love to Cheat

Shortcuts are decision rules that perform well on standard benchmarks but fail to transfer to more challenging testing conditions. Shortcut opportunities come in many flavors and are ubiquitous across datasets and application domains.

We here discuss how shortcut learning unifies many of deep learning’s problems and what we can do to better understand and mitigate shortcut learning.