Research map / Label-efficient and robust learning / Domain adaptation and generalization
Label Efficient Learning of Transferable Representations acrosss Domains and Tasks
Zelun Luo, Yuliang Zou, Judy Hoffman, Li F Fei-Fei
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 |
|---|---|---|---|
| add the full multi-layer domain and semantic transfer objectives | fine-tuning from SVHN to MNIST · MNIST 5-9 test, k=2 | accuracy: 0.612 ± 0.020 -> 0.917 ± 0.007 | Improved |
| add multi-layer adversarial domain adaptation without the full semantic transfer objective (ablation) | fine-tuning from SVHN to MNIST · MNIST 5-9 test, k=2 | accuracy: 0.612 ± 0.020 -> 0.702 ± 0.020 | Improved |
| add semantic transfer to fine-tuning plus adversarial adaptation (ablation) | SVHN-to-MNIST transfer · MNIST 5-9 test, k=2 | accuracy: 0.702 ± 0.020 -> 0.917 ± 0.007 | Improved |
| use the full method rather than fine-tuning | ImageNet-to-UCF-101 transfer · UCF-101, k=3 | per-video accuracy: 0.406±0.015 -> 0.467±0.007 | Improved |
| use the full method rather than fine-tuning | ImageNet-to-UCF-101 transfer · UCF-101, k=5 | per-video accuracy: 0.523±0.010 -> 0.545±0.014 | Improved |
| use the full method rather than fine-tuning | ImageNet-to-UCF-101 transfer · UCF-101, k=10 | per-video accuracy: 0.568±0.042 -> 0.620±0.005 | Improved |
| use the full method rather than fine-tuning | ImageNet-to-UCF-101 transfer · UCF-101, k=5 | per-frame accuracy: 0.486±0.012 -> 0.459±0.013 | Hurt |
| use the full method rather than fine-tuning | ImageNet-to-UCF-101 transfer · UCF-101, k=10 | per-frame accuracy: 0.529±0.039 -> 0.523±0.002 | No clear effect |
| replace the competing adversarial adaptation approach with the proposed multi-layer discriminator (ablation) | unsupervised SVHN-to-MNIST adaptation · MNIST test | accuracy: 0.760 ± 0.018 -> 0.810 ± 0.003 | Improved |
| use domain adversarial learning rather than source-only training | unsupervised SVHN-to-MNIST adaptation · MNIST test | accuracy: 0.601 ± 0.011 -> 0.810 ± 0.003 | Improved |
| replace the full method with target-only training from scratch | SVHN-to-MNIST transfer · MNIST 5-9 test, k=2 | accuracy: 0.917 ± 0.007 -> 0.642 ± 0.026 | Hurt |
Closest papers by meaning
- Unsupervised Video Domain Adaptation for Action Recognition: A Disentanglement Perspective (NeurIPS 2023 · similarity 0.87)
- Domain-Specific Priors and Meta Learning for Few-Shot First-Person Action Recognition (TPAMI 2021 · similarity 0.87)
- Transferable Representation Learning with Deep Adaptation Networks (TPAMI 2018 · similarity 0.87)
- A Broad Study on the Transferability of Visual Representations With Contrastive Learning (ICCV 2021 · similarity 0.87)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.86)
- Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training (CVPR 2024 · similarity 0.85)
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.
