Domain Expansion: A Latent Space Construction Framework for Multi-Task Learning

Chi-Yao Huang, Khoa Vo, Aayush Atul Verma, Duo Lu, Yezhou Yang · 2026

Read Domain Expansion: A Latent Space Construction Framework for Multi-Task Learning by Chi-Yao Huang, Khoa Vo, Aayush Atul Verma, Duo Lu, Yezhou Yang (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-2601-20069.txt.

Description

Chi-Yao Huang and four colleagues restructure how neural networks learn several tasks at once. A single network trained on multiple objectives faces conflicting gradients. The shared representation gets pulled in incompatible directions and ends up compromised for every task. The authors call this latent representation collapse.

Their framework, Domain Expansion, attacks the problem by restructuring the latent space itself. Giving each task more room should stop the tug of war. The network keeps a shared core while domains expand into their own territory. Multi-task systems power translation, perception, and recommendation products. A structural fix, rather than another weighting trick, could generalise across them. The work reframes a training nuisance as a geometry problem.

Catalog record

CREATORChi-Yao Huang, Khoa Vo, Aayush Atul Verma, Duo Lu, Yezhou Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-20069

Browse the full library · About the Tower of Babel