Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition

Minxue Tang, Yangyang Yu, [1mm] Aolin Ding, Maziyar Baran Pouyan, Taha Belkhouja, Yujia Bao · 2026

Read Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition by Minxue Tang, Yangyang Yu, [1mm] Aolin Ding, Maziyar Baran Pouyan, Taha Belkhouja, Yujia Bao (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-2602-19385.txt.

Description

Spotting implicit visual and textual patterns matters across modern AI uses, yet long-tail patterns challenge pretrained base models. Minxue Tang and five co-authors propose ADAMAB, a lean framework for tuning embeddings in few-shot pattern recognition. It trains lightweight, embedder-agnostic tuners above frozen embedding nets without touching their parameters, keeping compute cost low.

To reduce the need for large training sets, ADAMAB adds adaptive data boosts driven by a multi-armed bandit method. Using an adjusted upper confidence bound rule, it cuts gradient shifting and offers a theory-backed guarantee of convergence in few-shot training. Multimodal tests show up to 40 percent accuracy gain when training with fewer than five initial samples per class.

Catalog record

CREATORMinxue Tang, Yangyang Yu, [1mm] Aolin Ding, Maziyar Baran Pouyan, Taha Belkhouja, Yujia Bao
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-19385

Browse the full library · About the Tower of Babel