Research map / Architectures and efficiency / RNNs and normalization
Task Switching Network for Multi-Task Learning
Guolei Sun, Thomas Probst, Danda Pani Paudel, Nikola Popović, Menelaos Kanakis, Jagruti Patel, Dengxin Dai, Luc Van Gool
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 task embedding network (learned AdaIN coefficients) to task-specific INs (ablation) | TSN decoder · PASCAL-Context (5 tasks) | Delta_m %: 1.43 -> 0.30 | Improved |
| use instance normalization instead of batch normalization for task-specific affine parameters (ablation) | TSN decoder · PASCAL-Context (5 tasks) | Delta_m %: 4.38 -> 1.43 | Improved |
| replace multi-decoder with single shared conditional decoder (INs+TE) | TSN · PASCAL-Context (5 tasks) | Delta_m %: 4.32 -> 0.30 | Improved |
| increase orthogonal task-code dimension from 100 to 250 (ablation) | TSN task embedding · PASCAL-Context (5 tasks) | Delta_m %: 0.30 -> 0.75 | Hurt |
| use Gaussian instead of orthogonal task codes at d=100 (ablation) | TSN task embedding · PASCAL-Context (5 tasks) | Delta_m %: 0.30 -> 1.22 | Hurt |
| use Gaussian instead of orthogonal task codes at d=50 (ablation) | TSN task embedding · PASCAL-Context (5 tasks) | Delta_m %: 0.32 -> 0.30 | No clear effect |
| scale backbone from ResNet-18 to ResNet-34 (ablation) | TSN · PASCAL-Context (5 tasks) | Delta_m %: 0.30 -> 0.99 | Hurt |
| scale backbone from ResNet-18 to ResNet-101 (ablation) | TSN · PASCAL-Context (5 tasks) | Delta_m %: 0.30 -> 0.93 | Hurt |
| compare TSN with RCM on PASCAL-Context | TSN · PASCAL-Context (5 tasks) | Delta_m %: 3.62 -> 0.30 | Improved |
| compare TSN with RCM on NYUD | TSN · NYUD (4 tasks) | Delta_m %: 6.14 -> 0.03 | Improved |
| convert TSN to task-conditioned network by precomputing AdaIN parameters | TSN · PASCAL-Context (5 tasks) | # params: 18.3M -> 17.7M | Improved |
| share one prediction head across tasks instead of separate heads (ablation) | TSN · PASCAL-Context (5 tasks) | accuracy | No clear effect |
Closest papers by meaning
- MTI-Net: Multi-Scale Task Interaction Networks for Multi-Task Learning (ECCV 2020 · similarity 0.88)
- Controllable Dynamic Multi-Task Architectures (CVPR 2022 · similarity 0.88)
- Multi-Task Recurrent Modular Networks (AAAI 2021 · similarity 0.88)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.87)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.86)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 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.
