AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Achievement-Based Training Progress Balancing for Multi-Task Learning

Hayoung Yun, Hanjoo Cho

ICCV 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning · 13 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add achievement-based task weight to DTP (ablation)DTP · NYU v2AccMTL: 0.3660 -> 0.3745Improved
use weighted geometric mean of task losses instead of weighted sum (ablation)DTP + achievement weight · NYU v2AccMTL: 0.3745 -> 0.3847Improved
add achievement-based weights to PCGrad (ablation)PCGrad · NYU v2AccMTL: 0.3558 -> 0.3662Improved
add achievement-based weights to CAGrad (ablation)CAGrad · NYU v2AccMTL: 0.3556 -> 0.3653Improved
replace arithmetic weighted sum with weighted geometric mean in DWA (ablation)DWA · NYU v2AccMTL: 0.3676 -> 0.3811Improved
replace arithmetic weighted sum with weighted geometric mean in RLW (ablation)RLW · NYU v2AccMTL: 0.3673 -> 0.3774Improved
replace arithmetic weighted sum with weighted geometric mean in DTP (ablation)DTP · NYU v2AccMTL: 0.3660 -> 0.3800Improved
use gradient-based methods (MGDA, PCGrad, CAGrad) on partially annotated datamulti-task loss · VOC+NYUAccMTL: 0.8642 -> 0.7621 (MGDA vs Uniform)Hurt
use hard pseudo labels (MuST) instead of partially annotated labelsAMTL · VOC+NYUAccMTL: 0.8784 -> 0.8431Hurt
add single-task teacher distillation (KD-MTL) to multi-task trainingAMTL · VOC+NYUAccMTL: 0.8784 -> 0.8742Hurt
apply any multi-task loss vs single-task baseline with surface normal taskmulti-task losses · NYU v2AccMTL: 0.3989 -> 0.3847 (AMTL)Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy