Research map / Label-efficient and robust learning / Domain adaptation and generalization
Universal Representations: A Unified Look at Multiple Task and Domain Learning
Weihong Li, Xialei Liu, Hakan Bilen
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 |
|---|---|---|---|
| replace uniform-weight MTL with feature distillation into universal network | SegNet MTL · NYU-v2 | Delta MTL (%): -1.13 -> +10.95 | Improved |
| add feature distillation with adapters | MTAN MTL · NYU-v2 | Delta MTL (%): +1.65 -> +8.85 | Improved |
| add feature distillation | SegNet MTL · Cityscapes | Delta MTL (%): -0.74 -> +2.21 | Improved |
| distill predictions (BAM) instead of features (ablation) | SegNet MTL · NYU-v2 | Delta MTL (%): -1.13 -> -4.47 | Hurt |
| add feature distillation on top of Uncertainty loss balancing (ablation) | SegNet MTL · NYU-v2 | Delta MTL (%): +0.72 -> +12.80 | Improved |
| add feature distillation on top of IMTL-H (ablation) | SegNet MTL · NYU-v2 | Delta MTL (%): +3.11 -> +12.41 | Improved |
| add loss balancing (Uncertainty) to feature distillation vs uniform weights (ablation) | Ours SegNet · NYU-v2 | Delta MTL (%): +10.95 -> +12.80 | Improved |
| add feature distillation to PAD-Net | HRNet-18 PAD-Net · NYU-v2 | Delta MTL (%): -1.85 -> +0.28 | Improved |
| add feature distillation to MTI-Net | HRNet-18 MTI-Net · NYU-v2 | Delta MTL (%): +1.72 -> +4.11 | Improved |
| add feature distillation to MTI-Net | HRNet-18 MTI-Net · Cityscapes | Delta MTL (%): -0.17 -> +0.77 | Improved |
| add representation distillation to multi-domain learning (ablation) | ResNet-26 Visual Decathlon · Visual Decathlon | Delta MDL (%): +0.75 -> +4.00 | Improved |
| add residual adapters to universal network | ResNet-26 Visual Decathlon · Visual Decathlon | avg accuracy (%): 79.25 -> 80.52 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MTI-Net — MTI-Net: Multi-Scale Task Interaction Networks for Multi-Task Learning (ECCV 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Universal Representation Learning From Multiple Domains for Few-Shot Classification (ICCV 2021 · similarity 0.90)
- Learning More Universal Representations for Transfer-Learning (TPAMI 2019 · similarity 0.89)
- Learning multiple visual domains with residual adapters (NeurIPS 2017 · similarity 0.89)
- Characterizing Universal Object Representations Across Vision Models (NeurIPS 2026 · similarity 0.86)
- Towards Universal Dataset Distillation via Task-Driven Diffusion (CVPR 2025 · similarity 0.85)
- UDON: Universal Dynamic Online distillatioN for generic image representations (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.
