Research map / Architectures and efficiency / Pruning and efficient tuning
K for the Price of 1: Parameter-efficient Multi-task and Transfer 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 |
|---|---|---|---|
| Add the scale-and-bias patch to last-layer fine-tuning (ablation) | InceptionV3 transfer learning · Flowers-102 | accuracy: 84.5 -> 90.4 | Helped |
| Add the scale-and-bias patch to last-layer fine-tuning (ablation) | InceptionV3 transfer learning · Cars | accuracy: 55 -> 81 | Helped |
| Add the scale-and-bias patch to last-layer fine-tuning (ablation) | InceptionV3 transfer learning · Aircraft | accuracy: 45.9 -> 70.7 | Helped |
| Add the last layer to scale-and-bias-only tuning (ablation) | InceptionV3 transfer learning with a random last layer · Flowers-102 | accuracy: 79.5 -> 90.4 | Helped |
| Add the last layer to scale-and-bias-only tuning (ablation) | InceptionV3 transfer learning with a random last layer · Cars | accuracy: 33 -> 81 | Helped |
| Add the last layer to scale-and-bias-only tuning (ablation) | InceptionV3 transfer learning with a random last layer · Aircraft | accuracy: 52.3 -> 70.7 | Helped |
| Add the scale-and-bias patch to last-layer fine-tuning (ablation) | COCO SSD to ImageNet classification · ImageNet | Top1 accuracy: 29.2% -> 47.8% | Helped |
| Train a task-specific scale-and-bias patch and last layer instead of separate logits only (ablation) | MobileNetV2 multi-task learning · ImageNet | accuracy: 64.4% -> 70.2% | Helped |
| Train a task-specific scale-and-bias patch and last layer instead of separate logits only (ablation) | MobileNetV2 multi-task learning · Places-365 | accuracy: 51.4% -> 54.3% | Helped |
| Use a task-specific scale-and-bias patch rather than sharing all parameters (ablation) | MobileNetV2 resolution-adapted multi-task learning · ImageNet 128 x 128 | accuracy: 62.4% -> 66.3% | Helped |
| Use a task-specific scale-and-bias patch rather than sharing all parameters (ablation) | MobileNetV2 resolution-adapted multi-task learning · ImageNet 224 x 224 | accuracy: 70.6% -> 71.8% | Helped |
| Fine-tune a scale-and-bias patch and last layer instead of only the last layer (ablation) | COCO SSD to ImageNet classification · ImageNet | Top1 accuracy: 29.2% -> 47.8% | Helped |
Closest papers by meaning
- Borrowing Treasures From the Wealthy: Deep Transfer Learning Through Selective Joint Fine-Tuning (CVPR 2017 · similarity 0.88)
- Meta-Transfer Learning for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- A Mathematical Framework for Quantifying Transferability in Multi-source Transfer Learning (NeurIPS 2021 · similarity 0.88)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.86)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.86)
- Features are fate: a theory of transfer learning in high-dimensional regression (ICML 2025 · 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.