Research map / Learning theory and algorithms / Clustering
Efficient Task Grouping Through Sample-Wise Optimisation Landscape Analysis
Anshul Thakur, Yichen Huang, Soheila Molaei, Yujiang Wang, David Andrew Clifton
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 |
|---|---|---|---|
| use SCA informed task grouping instead of random grouping | Reptile joint training · Image classification (2 splits) | balanced accuracy average: 85.05 -> 86.17 | Improved |
| use SCA informed task grouping instead of random grouping | Reptile joint training · MIMIC-III Waveform ECG (2 splits) | AUROC average: 0.62 -> 0.65 | Improved |
| use SCA instead of TAG for task grouping | Reptile joint training · MIMIC-III Waveform ECG (2 splits) | AUROC average: 0.643 -> 0.65 | Improved |
| use SCA instead of TAG for task grouping | Reptile joint training · MIMIC-III Waveform ECG (3 splits) | AUROC average: 0.657 -> 0.657 | No clear effect |
| use SCA instead of TAG for task grouping | ResNet-15 MTL · Celeb-A | mean absolute error: 49.66 -> 49.6 | No clear effect |
| use SCA instead of TAG for task grouping (training time) | ResNet-15 MTL · Celeb-A (2 splits) | training time (GPU hours): 5.4 -> 3.61 | Improved |
| use SCA instead of HOA for task grouping (training time) | ResNet-15 MTL · Celeb-A (2 splits) | training time (GPU hours): 57.6 -> 3.61 | Improved |
| use SCA two-group task grouping instead of standard MTL | Celeb-A MTL · Celeb-A | relative improvement: 45% | Improved |
| use SCA two-group task grouping instead of Nash MTL | Celeb-A MTL · Celeb-A | relative improvement: 23.1% | Improved |
| use informed grouping (SCA, TAG, HOA) instead of random grouping | patient care Reptile · MIMIC-III patient care | AUROC average: 0.824 -> 0.845 | Improved |
| increase number of task groupings | all grouping methods · QM9 (2 to 7 splits) | mean absolute error | No clear effect |
| increase samples for SCA from 100 to 250, 500, 1000, 2000, 4000 (ablation) | SCA scores · image and patient care tasks | relative task similarity | No clear effect |
Closest papers by meaning
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.91)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.89)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.89)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.86)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.85)
- GO4Align: Group Optimization for Multi-Task Alignment (NeurIPS 2024 · similarity 0.85)
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.
