SpurAudio: A Benchmark for Studying Shortcut Learning in Few-Shot Audio Classification

Giries Abu Ayoub Department of Computer ScienceUniversity of Haifa [email protected] Morad Tukan Independent Researcher [email protected] Loay MualemUniversity of Stuttgart, Germany · 2026

Read SpurAudio: A Benchmark for Studying Shortcut Learning in Few-Shot Audio Classification by Giries Abu Ayoub Department of Computer ScienceUniversity of Haifa [email protected] Morad Tukan Independent Researcher [email protected] Loay MualemUniversity of Stuttgart, Germany (2026) 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/arxiv-2605-13672.txt.

Description

Giries Abu Ayoub, Morad Tukan and Loay Mualem introduce SpurAudio, a benchmark for studying shortcut learning in few-shot audio sorting. Few-shot sorting assumes target concepts are free of setting cues. SpurAudio uses the natural split of front events and background settings to enable staged multi-level tests of setting shifts across support and query sets. Many top few-shot methods degrade badly when background links break, even large trained audio base models. Methods close under standard tests can differ much in feeling for false links.

Catalog record

CREATORGiries Abu Ayoub Department of Computer ScienceUniversity of Haifa [email protected] Morad Tukan Independent Researcher [email protected] Loay MualemUniversity of Stuttgart, Germany
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE120231
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.13672

Browse the full library · About the Tower of Babel