Research map / Label-efficient and robust learning / Recognition and long-tail
CORE-MTL: Rethinking Gradient Balancing via Causal Orthogonal Representations
Chengfeng Wu, Tao Zou, Yanru Wu, Jingge Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add dual-stream semantic-residual backbone (ablation) | Vanilla MTL · NYUv2 | mIoU: 0.5249 -> 0.5352 | Improved |
| add reconstruction grounding to dual stream (ablation) | DS · NYUv2 | mIoU: 0.5352 -> 0.5424 | Improved |
| add CKA independence loss (ablation) | DS+Grounding · NYUv2 | mIoU: 0.5424 -> 0.5547 | Improved |
| add counterfactual augmentation (ablation) | DS+Grounding+CKA · NYUv2 | mIoU: 0.5547 -> 0.5693 | Improved |
| add CKA independence loss (ablation) | DS+Grounding · NYUv2 surface normal | 11.25 deg accuracy: 0.3517 -> 0.3165 | Hurt |
| use physics-based hard grounding decoder instead of generic soft decoder (ablation) | CORE-MTL · NYUv2 | mIoU: 0.5652 -> 0.5693 | Improved |
| use physics-based hard grounding decoder instead of generic soft decoder (ablation) | CORE-MTL · NYUv2 depth | Rel Err: 0.1427 -> 0.1466 | Hurt |
| set lambda_CKA to 1.0 instead of 0 (ablation) | CORE-MTL · NYUv2 | mIoU: 0.5636 -> 0.5694 | Improved |
| raise lambda_CKA to 100.0 from 1.0 (ablation) | CORE-MTL · NYUv2 | mIoU: 0.5694 -> 0.5551 | Hurt |
| use CORE-MTL instead of RepMTL | CORE-MTL · Cityscapes depth Rel Err | Rel Err: 44.3230 -> 19.6088 | Improved |
| use CORE-MTL instead of Equal Weighting | CORE-MTL · Cityscapes-C | Rel Err: 58.75 -> 38.10 | Improved |
| use CORE-MTL instead of Equal Weighting on 10-task CelebA | CORE-MTL · CelebA 10 tasks | Avg accuracy (%): 88.52 -> 89.39 | Improved |
Closest papers by meaning
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.87)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.87)
- NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective (NeurIPS 2025 · similarity 0.86)
- Joint Scheduling of Causal Prompts and Tasks for Multi-Task Learning (CVPR 2025 · similarity 0.86)
- Vision-Language Models Create Cross-Modal Task Representations (ICML 2025 · similarity 0.86)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.86)
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.
