Research map / Label-efficient and robust learning / Recognition and long-tail
ForkMerge: Mitigating Negative Transfer in Auxiliary-Task Learning
Junguang Jiang, Baixu Chen, Junwei Pan, Ximei Wang, Dapeng Liu, jie jiang, Mingsheng Long
This paper tried a common change: Add a reconstruction / auxiliary loss
add auxiliary task with gradient conflicts · DomainNet Painting/Quickdraw · transfer gain No clear effect
What 987 accepted papers report for this change
- 882 improved
- 41 no clear effect
- 20 mixed
- 44 hurt
- OPTION: Optimal Transport–Guided Flow Matching for Incomplete and Unaligned Multi-View Clustering ICML 2026 · add reconstruction loss L_rec · Scene-15 (aligned, 70% missing) · ACC: 32.4 -> 37.5Improved
- Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR ICML 2026 · add R-probe auxiliary loss · OCRBench · score: 714.0 -> 721.0Improved
- Rethinking Time-Series Imputation as Conditional Inference along Temporal Evolution ICML 2026 · add reconstruction loss on observed entries (joint vs missing-only supervision) · ETTh1 10% missing · Masked MAE: 0.2002 -> 0.1658Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace equal-weight joint training with ForkMerge | joint training · DomainNet Quickdraw | accuracy: 50.8 -> 73.0 | Improved |
| replace equal weighting with ForkMerge | DomainNet joint training · DomainNet | Delta_m: -9.62% -> +2.00% | Improved |
| ForkMerge vs single-task learning | ResNet-101 · DomainNet | average accuracy: 69.8 -> 71.1 | Improved |
| pretrain on all tasks then finetune (Post-train) | STL · DomainNet | average accuracy: 69.8 -> 70.4 | Improved |
| replace EW with ForkMerge | DeepLabV3+ · NYUv2 | Delta_m: 0.30% -> 4.03% | Improved |
| add task selection with 3 branches (ForkMerge++) (ablation) | ForkMerge · NYUv2 | Delta_m: 4.03% -> 4.59% | Improved |
| increase branches from 2 to 3 (ablation) | ForkMerge · DomainNet | Delta_m: 2.0% -> 3.0% | Improved |
| increase branches from 3 to 6 (ablation) | ForkMerge · DomainNet | Delta_m: 3.0% -> 3.6% | Improved |
| gain per added branch as branches increase (ablation) | ForkMerge · DomainNet | Delta_m/(B-1): 2.0% -> 0.7% | Hurt |
| ForkMerge vs STL on AliExpress | ESMM · AliExpress | average AUC: 0.7963 -> 0.8067 | Improved |
| equal weighting vs STL | ESMM · AliExpress | Delta_m: -0.39% | Hurt |
| ForkMerge replaces fixed-weight grid search for S4L | S4L · CIFAR-10 4000 labels | test error: 13.8 -> 13.1 | Improved |
Closest papers by meaning
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.87)
- Mitigating Negative Transfer in Multi-Task Learning with Exponential Moving Average Loss Weighting Strategies (Student Abstract) (AAAI 2023 · similarity 0.87)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 2024 · similarity 0.86)
- Aux-NAS: Exploiting Auxiliary Labels with Negligibly Extra Inference Cost (ICLR 2024 · similarity 0.84)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.84)
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.
