Research map / Label-efficient and robust learning / Recognition and long-tail
Improving Predictors via Combination Across Diverse Task Categories
Kwang In Kim
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 HSIC structure-preservation regularizer to MMD-based chart map learning (ablation) | predictor combination chart maps · visual attribute ranking across 5 datasets (174 targets) | % target attributes significantly improved over MMD-only: 55.75% -> 71.84% (vs f^I) | Improved |
| Align target data to reference domain via MMD-HSIC chart maps instead of identity maps (ablation) | predictor combination · datasets sharing feature space (AWA2, CUB, PubFig ResNet, Shoes ResNet, OSR ResNet) | % attributes significantly better than Id: 60.54% of all targets significantly improved over Id | Improved |
| Replace MMD-only alignment with image translation (CycleGAN-style) for chart maps (ablation) | predictor combination · ResNet-feature datasets | % attributes significantly improved by Ours over IT: IT performance comparable to MMD; Ours beats IT on 62.59% of targets | No clear effect |
| Use identity chart maps (evaluating references directly on target data) for predictor combination (ablation) | predictor combination · AWA2 and other datasets sharing domain | ranking of method: Id ranked second best only on AWA2 (after Ours) | Hurt |
| Entropy regularizer on reference weights in marginal likelihood (ablation) | predictor combination (OSR dataset preliminary) · OSR dataset | average accuracy of final predictors: around 5.4% improvement (preliminary) | Improved |
| Proposed predictor combination across heterogeneous task categories (MMD-HSIC alignment) vs initial SVM predictor | SVM rank predictors · AWA2 / CUB / PubFig / Shoes / OSR attribute ranking | number of targets significantly better than f^I: 71.84% of 174 targets significantly improved; none significantly degraded | Improved |
Closest papers by meaning
- Combining Task Predictors via Enhancing Joint Predictability (ECCV 2020 · similarity 0.92)
- Joint Manifold Diffusion for Combining Predictions on Decoupled Observations (CVPR 2019 · similarity 0.89)
- Identifying and Mitigating Spurious Correlation in Multi-Task Learning (CVPR 2025 · similarity 0.82)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.82)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.82)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 2024 · similarity 0.82)
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.
