Semantics-Aware Generative Latent Data Augmentation for Learning in Low-Resource Domains

Jaesung Bae, Minje Kim · 2026

Read Semantics-Aware Generative Latent Data Augmentation for Learning in Low-Resource Domains by Jaesung Bae, Minje Kim (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-2602-02841.txt.

Description

Jaesung Bae and 1 co-author tackle learning when labeled data is scarce in a 2026 paper. Their framework, GeLDA, creates extra training examples in a compressed latent space. A latent space is the compact set of features a foundation model extracts from raw inputs.

GeLDA uses conditional diffusion models to generate those examples. Extra feature vectors guide generation toward the right class or subdomain. Two large tests follow. One covers speech emotion recognition in a new language with no training examples. The other covers image classification with rare classes. In both, GeLDA improved results, including a new best score for rare image classes.

Catalog record

CREATORJaesung Bae, Minje Kim
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE90 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02841

Browse the full library · About the Tower of Babel