NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective
Xiaohan Qin, Xiaoxing Wang, Ning Liao, Junchi YanSchool of CS & School of AI, Shanghai Jiao Tong UniversityShanghai Innovation Institute · 2025
Read NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective by Xiaohan Qin, Xiaoxing Wang, Ning Liao, Junchi YanSchool of CS & School of AI, Shanghai Jiao Tong UniversityShanghai Innovation Institute (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-2510-18258.txt.
Description
Xiaohan Qin, with Xiaoxing Wang and two co-authors, tackles uneven learning across tasks in a 2025 paper. Multi-task learning trains one model on several jobs at once, sharing what it learns. Some tasks learn fast and crowd out slower ones, a problem called task imbalance.
The proposed method, NTKMTL, studies training through Neural Tangent Kernel theory. That theory tracks how a network's outputs shift as its settings change during training. By comparing the speeds of different tasks, the method rebalances them during learning. A faster variant shares representations across tasks to save time. Tests across supervised and reinforcement benchmarks reached top results.
Catalog record
| CREATOR | Xiaohan Qin, Xiaoxing Wang, Ning Liao, Junchi YanSchool of CS & School of AI, Shanghai Jiao Tong UniversityShanghai Innovation Institute |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-18258