Research map / Label-efficient and robust learning / Recognition and long-tail
Borrowing Treasures From the Wealthy: Deep Transfer Learning Through Selective Joint Fine-Tuning
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 |
|---|---|---|---|
| Add selectively retrieved source images and jointly fine-tune shared convolutional layers | ResNet-152 fine-tuning · Stanford Dogs 120 | mean Acc(%): 80.4 -> 90.2 | Helped |
| Add selectively retrieved source images and jointly fine-tune shared convolutional layers | ResNet-152 fine-tuning · Oxford Flowers 102 | mean Acc(%): 92.3 -> 94.7 | Helped |
| Add selectively retrieved source images and jointly fine-tune shared convolutional layers | ResNet-152 fine-tuning · Caltech 256 | mean Acc(%): 76.4±0.1 -> 80.5±0.3; 81.2±0.2 -> 83.8±0.5; 83.5±0.2 -> 87.0±0.1; 86.4±0.3 -> 89.1±0.2 | Helped |
| Use Places as the source domain for selective joint fine-tuning | ResNet-152 · MIT Indoor 67 | mean Acc(%): 81.7 -> 85.8 | Helped |
| Use ImageNet and Places together instead of Places alone as source data | Selective joint fine-tuning · MIT Indoor 67 | mean Acc(%): 85.8 -> 85.5 | Hurt |
| Fuse outputs from the Places-source and hybrid-source models | Selective joint fine-tuning · MIT Indoor 67 | mean Acc(%): 85.8 -> 86.9 | Helped |
| Use all source images rather than a selected subset (ablation) | Selective joint fine-tuning · Stanford Dogs 120 | mean Acc(%): -4.6% | Hurt |
| Use all source images rather than a selected subset (ablation) | Selective joint fine-tuning · Oxford Flowers 102 | mean Acc(%): -1.3% | Hurt |
| Choose the same number of source images randomly instead of retrieving similar images (ablation) | Selective joint fine-tuning · Stanford Dogs 120 | mean Acc(%): -4.7% | Hurt |
| Choose the same number of source images randomly instead of retrieving similar images (ablation) | Selective joint fine-tuning · Oxford Flowers 102 | mean Acc(%): -1.5% | Hurt |
| Adaptively increase retrieved neighbors for hard target samples (ablation) | Selective joint fine-tuning · Stanford Dogs 120 | mean Acc(%): +1.9% | Helped |
| Adaptively increase retrieved neighbors for hard target samples (ablation) | Selective joint fine-tuning · Oxford Flowers 102 | mean Acc(%): +0.5% | Helped |
Closest papers by meaning
- AdaFilter: Adaptive Filter Fine-Tuning for Deep Transfer Learning (AAAI 2020 · similarity 0.90)
- Co-Tuning for Transfer Learning (NeurIPS 2020 · similarity 0.90)
- K for the Price of 1: Parameter-efficient Multi-task and Transfer Learning (ICLR 2019 · similarity 0.88)
- Explicit Inductive Bias for Transfer Learning with Convolutional Networks (ICML 2018 · similarity 0.87)
- Growing a Brain: Fine-Tuning by Increasing Model Capacity (CVPR 2017 · similarity 0.87)
- Transferring Knowledge From Large Foundation Models to Small Downstream Models (ICML 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.