Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.20069 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908858201407488 |
|---|---|
| author | Huang, Chi-Yao Vo, Khoa Verma, Aayush Atul Lu, Duo Yang, Yezhou |
| author_facet | Huang, Chi-Yao Vo, Khoa Verma, Aayush Atul Lu, Duo Yang, Yezhou |
| contents | Training a single network with multiple objectives often leads to conflicting gradients that degrade shared representations, forcing them into a compromised state that is suboptimal for any single task--a problem we term latent representation collapse. We introduce Domain Expansion, a framework that prevents these conflicts by restructuring the latent space itself. Our framework uses a novel orthogonal pooling mechanism to construct a latent space where each objective is assigned to a mutually orthogonal subspace. We validate our approach across diverse benchmarks--including ShapeNet, MPIIGaze, and Rotated MNIST--on challenging multi-objective problems combining classification with pose and gaze estimation. Our experiments demonstrate that this structure not only prevents collapse but also yields an explicit, interpretable, and compositional latent space where concepts can be directly manipulated. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_20069 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Domain Expansion: A Latent Space Construction Framework for Multi-Task Learning Huang, Chi-Yao Vo, Khoa Verma, Aayush Atul Lu, Duo Yang, Yezhou Machine Learning Training a single network with multiple objectives often leads to conflicting gradients that degrade shared representations, forcing them into a compromised state that is suboptimal for any single task--a problem we term latent representation collapse. We introduce Domain Expansion, a framework that prevents these conflicts by restructuring the latent space itself. Our framework uses a novel orthogonal pooling mechanism to construct a latent space where each objective is assigned to a mutually orthogonal subspace. We validate our approach across diverse benchmarks--including ShapeNet, MPIIGaze, and Rotated MNIST--on challenging multi-objective problems combining classification with pose and gaze estimation. Our experiments demonstrate that this structure not only prevents collapse but also yields an explicit, interpretable, and compositional latent space where concepts can be directly manipulated. |
| title | Domain Expansion: A Latent Space Construction Framework for Multi-Task Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2601.20069 |