AI Paper Atlas

Research map / Optimization / Optimizers

Learning Task-Preferred Inference Routes for Gradient De-Conflict in Multi-Output DNNs

Yi Sun, Xiaochang Hu, Xin Xu, Jian Li, Yifei Shi, Ling‐Li Zeng

TPAMI 2025 · Optimization › Optimizers · approach: Constrained and non-backprop training · 4 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
train with DR-MGF instead of plain SGDVgg-SDN · CIFAR100average accuracy: 66.34 -> 69.25Improved
train with DR-MGF instead of plain SGDVgg-SDN · CIFAR10average accuracy: 88.15 -> 90.50Improved
train with DR-MGF instead of plain SGDResnet-SDN · CIFAR100average accuracy: 59.29 -> 61.84Improved
train with DR-MGF instead of plain SGDMSDnet · ImageNetaverage accuracy: 66.72 -> 67.35Improved
add DR-MGF on top of knowledge distillationMSDnet · CIFAR100average accuracy: 72.97 -> 74.13Improved
use DR-MGF instead of Nash-MTLSegNet-MTAN · NYUv2Delta m%: -4.04 -> -8.39Improved
apply Nash-MTL instead of plain SGDVgg-SDN · CIFAR100average accuracy: 66.34 -> 64.29Hurt
apply Cagrad instead of plain SGDVgg-SDN · CIFAR100 exitsaccuracy by exit: Exit-1 44.42 -> 53.08; Exit-7 71.76 -> 71.43Mixed
apply DR-MGF instead of plain SGD at the deepest exitResnet-SDN · CIFAR100 Exit-7accuracy: 70.02 -> 69.42Hurt
add network disentanglement to meta-weighted gradient fusion (ablation)Meta-GF · CIFAR100average accuracy: 68.51 -> 69.25Improved
add meta-weighted gradient fusion to disentanglement with average fusion (ablation)DR-avgF · CIFAR100convergence accuracyImproved
train baselines for 700 epochs to match DR-MGF training timeCagrad · CIFAR100 Vgg-SDNaverage accuracy: 67.60 -> 68.93No clear effect

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