Research map / Architectures and efficiency / RNNs and normalization
AdaShare: Learning What To Share For Efficient Deep Multi-Task Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| learn task-specific layer-sharing policies | AdaShare · NYU v2 3-Task | Δ_T: +8.9 vs. +2.7 for MTAN | Helped |
| learn task-specific layer-sharing policies | AdaShare · DomainNet 6 domains | accuracy: 57.6% -> 62.2% | Helped |
| learn task-specific layer-sharing policies | AdaShare · 10 text classification datasets | accuracy: 68.9% -> 76.1% | Helped |
| add curriculum training (ablation) | AdaShare · NYU v2 3-Task | Δ_T: +5.6 -> +8.9 | Helped |
| add sparsity regularization (ablation) | AdaShare · NYU v2 3-Task | Δ_T: +0.7 -> +8.9 | Helped |
| add sharing regularization (ablation) | AdaShare · NYU v2 3-Task | Δ_T: +5.6 -> +8.9 | Helped |
| add curriculum training (ablation) | AdaShare · CityScapes 2-Task | Δ_T: -0.5 -> +2.8 | Helped |
| add sparsity regularization (ablation) | AdaShare · CityScapes 2-Task | Δ_T: +1.7 -> +2.8 | Helped |
| add sharing regularization (ablation) | AdaShare · CityScapes 2-Task | Δ_T: +1.8 -> +2.8 | Helped |
| use instance-specific instead of task-specific policies (ablation) | AdaShare · NYU v2 3-Task | Δ_T: +8.9 vs. -6.9 | Hurt |
| replace Gumbel-Softmax with REINFORCE (ablation) | AdaShare · NYU v2 3-Task | Δ_T: +8.9 vs. -3.7 | Hurt |
| learn skipped-block locations rather than randomizing them at matched per-task skip counts (ablation) | AdaShare · NYU v2 3-Task | Δ_T: + 6.6 -> +8.9 | Helped |
Closest papers by meaning
- AdaTask: A Task-Aware Adaptive Learning Rate Approach to Multi-Task Learning (AAAI 2023 · similarity 0.88)
- Cross-Stitch Networks for Multi-Task Learning (CVPR 2016 · similarity 0.87)
- Modular Universal Reparameterization: Deep Multi-task Learning Across Diverse Domains (NeurIPS 2019 · similarity 0.86)
- Learning Sparse Sharing Architectures for Multiple Tasks (AAAI 2020 · similarity 0.86)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.84)
- DiTASK: Multi-Task Fine-Tuning with Diffeomorphic Transformations (CVPR 2025 · similarity 0.84)
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.