fastabx: A library for efficient computation of ABX discriminability

Maxime Poli, Emmanuel Chemla, Emmanuel Dupoux · 2025

Read fastabx: A library for efficient computation of ABX discriminability by Maxime Poli, Emmanuel Chemla, Emmanuel Dupoux (2025) 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-2505-02692.txt.

Description

Maxime Poli, Emmanuel Chemla, and Emmanuel Dupoux release fastabx, a fast Python library for building ABX discrimination tasks. The ABX test measures how cleanly a learned representation separates categories. Given two examples of different classes and a third of one of them, does the model place the third with its kin? Speech scientists use it to grade self-supervised models.

Adoption stalled for lack of good tools. The library constructs any ABX task variant and computes the needed distances efficiently, suiting rapid experiment cycles. The authors aim it beyond speech, at any field that learns representations. A shared, fast yardstick makes model comparisons honest.

Catalog record

CREATORMaxime Poli, Emmanuel Chemla, Emmanuel Dupoux
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-02692

Browse the full library · About the Tower of Babel