New framework CRLoMTL improves multi-task learning with low-rank matrices

Researchers have developed CRLoMTL, a novel framework designed to improve multi-task learning (MTL) in computer vision by addressing negative transfer. This method utilizes low-rank matrices to resolve conflicts between tasks, learning task-specific updates and integrating them through a shared-subspace fusion mechanism based on singular value decomposition. Experiments using Swin Transformer backbones on datasets like PASCAL-Context and NYUD-v2 demonstrated that CRLoMTL achieves superior accuracy with significantly fewer trainable parameters compared to single-task baselines. AI

IMPACT This research offers a more parameter-efficient approach to multi-task learning, potentially improving performance on complex vision tasks.

RANK_REASON The cluster contains a research paper detailing a new method for multi-task learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Celeba
• CRLoMTL
• LoRA+
• multi-task learning
• NYUD-v2
• PASCAL-Context
• singular value decomposition
• Swin Transformer

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →