Research map / Architectures and efficiency / RNNs and normalization
Modular Universal Reparameterization: Deep Multi-task Learning Across Diverse Domains
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 |
|---|---|---|---|
| Share MUiR modules across WRN and Stacked LSTM | Stacked LSTM · WikiText-2 | perplexity: 134.41 -> 130.63 | Helped |
| Share MUiR modules across WRN and DeepBind | DeepBind-256 · CRISPR binding prediction | MSE: 0.1540 -> 0.1461 | Helped |
| Share MUiR modules across LeNet and DeepBind | LeNet · CIFAR-10 | classification error: 21.08 -> 19.59 | Helped |
| Share modules with MUiR | LeNet · CIFAR-10 | classification error: 21.46 -> 20.51 | Helped |
| Share modules with MUiR | Stacked LSTM · WikiText-2 | perplexity: 135.03 -> 130.70 | Helped |
| Share modules with MUiR | DeepBind · CRISPR binding prediction | MSE: 0.1543 -> 0.1464 | Helped |
| Initialize module alignment hierarchically rather than disjointly (ablation) | MUiR with LeNet · CIFAR-10 | classification error: 20.51 -> 20.72 | Hurt |
| Initialize module alignment hierarchically rather than disjointly (ablation) | MUiR with Stacked LSTM · WikiText-2 | perplexity: 130.70 -> 128.94 | Helped |
| Set context size c to 0, removing location-specific contexts (ablation) | MUiR with Stacked LSTM · WikiText-2 | perplexity: 130.70 -> 144.52 | Hurt |
| Vary context size from c=4 to c=8 (ablation) | MUiR · WikiText-2 | perplexity: 130.70 -> 130.80 | No clear effect |
| Apply MUiR rather than single-task learning | AWD-LSTM with 8.8M parameters · WikiText-2 | perplexity: 73.64 -> 71.01 | Helped |
| Apply MUiR to the tuned AWD-LSTM setup rather than using the larger STL model | AWD-LSTM · WikiText-2 | perplexity: 69.94 -> 71.01 | Hurt |
Closest papers by meaning
- Twin-Merging: Dynamic Integration of Modular Expertise in Model Merging (NeurIPS 2024 · similarity 0.87)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 2024 · similarity 0.86)
- Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent (ICML 2025 · similarity 0.86)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.86)
- Customizable Combination of Parameter-Efficient Modules for Multi-Task Learning (ICLR 2024 · similarity 0.86)
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.