Unmasking Clever Hans predictors and assessing what machines really learn.

Lapuschkin S, Wäldchen S, Binder A, Montavon G, Samek W, Müller KR. · 2019

Read Unmasking Clever Hans predictors and assessing what machines really learn. by Lapuschkin S, Wäldchen S, Binder A, Montavon G, Samek W, Müller KR. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6411769.txt.

Description

Lapuschkin and co-authors probe what learning machines really learn, in Nature Communications. Top scores can hide lazy tricks, much like the horse Clever Hans seemed to count. The team used methods that explain a model's choices on vision and game tasks.

Some models used sound clues, while others leaned on shortcuts in the data. Standard accuracy scores could not tell these apart, they show. Their Spectral Relevance Analysis helps sort real strategy from trick. Such checks matter before models are trusted in medicine or transport. The work is a warning against reading accuracy as understanding.

Catalog record

CREATORLapuschkin S, Wäldchen S, Binder A, Montavon G, Samek W, Müller KR.
YEAR2019
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6411769

Browse the full library · About the Tower of Babel