notesum.ai
Published at November 27Proactive Gradient Conflict Mitigation in Multi-Task Learning: A Sparse Training Perspective
cs.LG
cs.AI
cs.CV
Released Date: November 27, 2024
Authors: Zhi Zhang1, Jiayi Shen1, Congfeng Cao1, Gaole Dai2, Shiji Zhou3, Qizhe Zhang2, Shanghang Zhang2, Ekaterina Shutova1
Aff.: 1ILLC, University of Amsterdam, Netherlands; 2State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University, China; 3Department of Automation, Tsinghua University, China

| Methods | Segmentation | Depth | Surface Normal | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| mIoU | Pix Acc | Abs Err | Rel Err | Angle Distance | Within | ||||||
| Mean | Median | 11.25 | 22.5 | 30 | |||||||
| STL | |||||||||||
| Joint Train | |||||||||||
| w/ ST | |||||||||||
| PCGrad | |||||||||||
| w/ ST | 80.33 | ||||||||||
| CAGrad | 0.3215 | 0.1305 | |||||||||
| w/ ST | |||||||||||
| GradDrop | |||||||||||
| w/ ST | |||||||||||
| MGDA | 12.61 | 72.35 | 80.87 | ||||||||
| w/ ST | 19.22 | 12.61 | 46.44 | ||||||||
| IMTL-G | 60.64 | ||||||||||
| w/ ST | |||||||||||
| NashMTL | |||||||||||
| w/ ST | |||||||||||