Research map / Architectures and efficiency / Neural architecture search
MTL-NAS: Task-Agnostic Neural Architecture Search Towards General-Purpose Multi-Task Learning
Check 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 |
|---|---|---|---|
| Search hierarchical, layerwise inter-task feature-fusion connections between fixed task-specific backbones | MTL-NAS · NYU v2, semantic segmentation and surface normal estimation | mIoU: 36.2 -> 37.6 | Helped |
| Search hierarchical, layerwise inter-task feature-fusion connections between fixed task-specific backbones | MTL-NAS · Taskonomy tiny split, object and scene classification | object Top-1 recognition rate: 34.1 -> 34.8 | Helped |
| Add minimum entropy regularization to DARTS-style search (ablation) | DARTS-style MTL-NAS search · NYU v2 | mIoU: 19.1 -> 37.7 | Helped |
| Add minimum entropy regularization to DARTS-style search (ablation) | DARTS-style MTL-NAS search · NYU v2 | surface normal mean error: 15.9 -> 12.7 | Helped |
| Use stochastic relaxation without minimum entropy regularization (ablation) | SNAS-style MTL-NAS search · NYU v2 | evaluation performance stability | Hurt |
| Initialize task-identical fusion weight w_TI to 1.0 rather than 0 (ablation) | MTL-NAS fusion layers · NYU v2 | mIoU: 30.1 -> 37.6 | Helped |
| Initialize task-identical fusion weight w_TI to 1.0 rather than 0 (ablation) | MTL-NAS fusion layers · NYU v2 | surface normal mean error: 16.9 -> 12.6 | Helped |
| Use the searched architecture instead of retaining all search-space edges in the supernet (ablation) | MTL-NAS · NYU v2 | mIoU: 36.1 -> 37.6 | Helped |
| Vary the learning rate of new inter-task layers from 1x to 10x (ablation) | MTL-NAS · NYU v2 | mIoU: 37.6 -> 37.5 | No clear effect |
| Use stochastic instead of deterministic discretization under minimum entropy regularization (ablation) | MTL-NAS with deterministic relaxation · NYU v2 | mIoU: 37.7 -> 37.9 | No clear effect |
Closest papers by meaning
- M-NAS: Meta Neural Architecture Search (AAAI 2020 · similarity 0.92)
- TransNAS-Bench-101: Improving Transferability and Generalizability of Cross-Task Neural Architecture Search (CVPR 2021 · similarity 0.91)
- Generic Neural Architecture Search via Regression (NeurIPS 2021 · similarity 0.90)
- Few-shot Task-agnostic Neural Architecture Search for Distilling Large Language Models (NeurIPS 2022 · similarity 0.89)
- Rapid Model Architecture Adaption for Meta-Learning (NeurIPS 2022 · similarity 0.89)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.88)
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.