Gradient-Based Multi-Objective Deep Learning: Algorithms, Theories, Applications, and Beyond

Weiyu Chen, Baijiong Lin, Xiaoyuan Zhang, Xi Lin, Han Zhao, Qingfu Zhang, James T. Kwok · 2025

Read Gradient-Based Multi-Objective Deep Learning: Algorithms, Theories, Applications, and Beyond by Weiyu Chen, Baijiong Lin, Xiaoyuan Zhang, Xi Lin, Han Zhao, Qingfu Zhang, James T. Kwok (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-2501-10945.txt.

Description

Weiyu Chen and six co-authors survey, in a 2025 preprint, gradient-based methods for teaching one network to serve many masters. Modern tasks often pull in opposite directions: accuracy against fairness, or many tasks against each other. Multi-objective optimization supplies the math of trade-offs. Its Pareto frontier collects the solutions where one goal can improve only at another's cost. Applying its gradient tools to deep networks brings cost, instability, and preference-handling troubles. The survey sorts the known algorithms by what they hand back.

Some return a single balanced answer. Others return a small set of spread-out trade-offs. A third family learns a whole continuum of solutions at once. The authors cover theory, applications, and shared resources. Open problems and promising directions close the paper. A public list of algorithms accompanies it, aimed at practitioners picking a method for multi-task or fairness-aware learning.

Catalog record

CREATORWeiyu Chen, Baijiong Lin, Xiaoyuan Zhang, Xi Lin, Han Zhao, Qingfu Zhang, James T. Kwok
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE101 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10945

Browse the full library · About the Tower of Babel