AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Resolving Token-Space Gradient Conflicts: Token Space Manipulation for Transformer-Based Multi-Task Learning

Wooseong Jeong, Kuk-Jin Yoon

ICCV 2025 · Architectures and efficiency › Pruning and efficient tuning · approach: Parameter-efficient transfer analysis · 2 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 token modulation (TM) to multi-task baseline (ablation)Baseline (MT) · NYUD-v2 SemsegmIoU: 34.13 -> 37.85Improved
add token expansion (TE) to multi-task baseline (ablation)Baseline (MT) · NYUD-v2 SemsegmIoU: 34.13 -> 37.25Improved
combine TM and TE (ablation)Baseline (MT) · NYUD-v2 SemsegmIoU: 34.13 -> 38.27Improved
combine TM and TE (ablation)Baseline (MT) · PASCAL-Context SemsegmIoU: 54.47 -> 66.18Improved
apply TM+TE to highest-conflict layers instead of randomly chosen layers (ablation)DTME-MTL · NYUD-v2delta_m (%): -2.966 -> +0.044Improved
apply TM+TE to highest-conflict layers instead of lowest-conflict layers (ablation)DTME-MTL · NYUD-v2delta_m (%): -6.167 -> +0.044Improved
assign TM to range-space and TE to null-space conflicts instead of swapped (ablation)DTME-MTL · NYUD-v2delta_m (%): -2.608 -> +0.044Improved
use affine modulator instead of batch norm (ablation)DTME-MTL · NYUD-v2 SemsegmIoU: 37.42 -> 38.27Improved
do not freeze backbone after expansion (ablation)DTME-MTL · NYUD-v2 SemsegmIoU: 34.80 -> 38.27Improved
add DTME-MTL to InvPTInvPT · PASCAL-Context SemsegmIoU: 79.03 -> 81.91Improved
add DTME-MTL to TaskPrompterTaskPrompter · PASCAL-Context SemsegmIoU: 80.89 -> 81.01Improved
DTME-MTL instead of Recon (parameter-level expansion)ViT-T MT · NYUD-v2 SemsegmIoU: 31.92 -> 38.27Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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