Research map / Label-efficient and robust learning / Domain adaptation and generalization
Auto-Transfer: Learning to Route Transferable Representations
Keerthiram Murugesan, Vijay Sadashivaiah, Ronny Luss, Karthikeyan Shanmugam, Pin-Yu Chen, Amit Dhurandhar
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 |
|---|---|---|---|
| route source representations to target layers with Auto-Transfer | Auto-Transfer vs SAD · CUB200 | accuracy: 68.29 -> 74.76 | Improved |
| route source representations to target layers with Auto-Transfer | Auto-Transfer vs SAD · Stanford Dogs | accuracy: 76.06 -> 86.16 | Improved |
| route source representations to target layers with Auto-Transfer | Auto-Transfer vs SAD · MIT67 | accuracy: 66.47 -> 75.86 | Improved |
| route source representations to target layers with Auto-Transfer | Auto-Transfer vs SAD · Stanford40 | accuracy: 63.08 -> 80.10 | Improved |
| use Auto-Transfer rather than L2T-ww | CIFAR100 with TinyImageNet source · CIFAR100 | accuracy: 70.96{$\pm$ 0.61} -> 72.48{$\pm$ 0.42} (full); 70.96{$\pm$ 0.61} -> 70.48{$\pm$ 0.25} (fixed); 70.96{$\pm$ 0.61} -> 70.8 | Mixed |
| use Auto-Transfer rather than L2T-ww | STL10 with TinyImageNet source · STL10 | accuracy: 78.31{$\pm$ 0.21} -> 82.09{$\pm$ 0.29} | Improved |
| use matched-architecture Auto-Transfer instead of finetuning | ResNet18 source and target · CUB200 | accuracy: 42.96{$\pm$ 1.45} -> 66.97{$\pm$ 1.38} | Improved |
| use matched-architecture Auto-Transfer instead of finetuning | ResNet18 source and target · Stanford Dogs | accuracy: 53.02{$\pm$ 3.57} -> 79.46{$\pm$ 1.05} | Improved |
| use matched-architecture Auto-Transfer instead of finetuning | ResNet18 source and target · MIT67 | accuracy: 47.93{$\pm$ 3.66} -> 69.54{$\pm$ 2.49} | Improved |
| use matched-architecture Auto-Transfer instead of finetuning | ResNet18 source and target · Stanford40 | accuracy: 34.40{$\pm$ 5.94} -> 75.07{$\pm$ 2.55} | Improved |
| fix selected layer pairs rather than using bandit routing during training (ablation) | Auto-Transfer · CUB200 | accuracy: 73.09 -> 75.15 | Improved |
| fix selected layer pairs rather than using bandit routing during training (ablation) | Auto-Transfer · Stanford Dogs | accuracy: 85.05 -> 86.40 | Improved |
Later papers that compared with this method
1 later accepted paper reports AutoTransfer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Beyond Model Base Retrieval: Weaving Knowledge to Master Fine-grained Neural Network Design (ICML 2026 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarially-Trained Deep Nets Transfer Better: Illustration on Image Classification (ICLR 2021 · similarity 0.88)
- Learning What and Where to Transfer (ICML 2019 · similarity 0.87)
- MineGAN: Effective Knowledge Transfer From GANs to Target Domains With Few Images (CVPR 2020 · similarity 0.87)
- Adversarial Training Helps Transfer Learning via Better Representations (NeurIPS 2021 · similarity 0.87)
- A Mathematical Framework for Quantifying Transferability in Multi-source Transfer Learning (NeurIPS 2021 · similarity 0.87)
- Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model (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.
