Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learngene: From Open-World to Your Learning Task
Qiu-Feng Wang, Xin Geng, Shu-Xia Lin, Shi-Yu Xia, Lei Qi, Ning Xu
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 |
|---|---|---|---|
| use learngene initialization and adaptation instead of DeepEMD | Learngene individual model · CIFAR100 open-world classes | accuracy: 39.85±0.41 -> 45.88±2.99 | Improved |
| use learngene initialization and adaptation instead of Fine-tuning | Learngene individual model · ImageNet100 open-world classes | accuracy: 40.72±1.36 -> 42.40±2.35 | Improved |
| use learngene initialization and adaptation instead of DeepEMD | Learngene individual model · CIFAR100 5-way 10-shot novel classes | accuracy: 58.03±1.85 -> 58.86±3.12 | Improved |
| use learngene initialization and adaptation instead of DeepEMD | Learngene individual model · CIFAR100 5-way 20-shot novel classes | accuracy: 63.69±2.05 -> 64.24±1.89 | Improved |
| use learngene initialization and adaptation instead of MatchingNet | Learngene individual model · ImageNet100 5-way 10-shot novel classes | accuracy: 51.82±2.58 -> 52.96±4.84 | Improved |
| use learngene initialization and adaptation instead of Fine-tuning | Learngene individual model · ImageNet100 5-way 20-shot novel classes | accuracy: 60.71±2.32 -> 60.86±4.48 | Improved |
| increase the retain-loss weight λ from 1000 to 5000 (ablation) | Learngene individual model · CIFAR100 5-way 10-shot novel classes | accuracy: 58.86±3.12 -> 58.21±2.58 | Hurt |
| increase the retain-loss weight λ from 1000 to 5000 (ablation) | Learngene individual model · CIFAR100 5-way 20-shot novel classes | accuracy: 64.24±1.89 -> 63.74±2.46 | Hurt |
| increase the retain-loss weight λ from 1000 to 5000 (ablation) | Learngene individual model · ImageNet100 5-way 10-shot novel classes | accuracy: 52.96±4.84 -> 52.53±4.74 | Hurt |
| increase the retain-loss weight λ from 1000 to 5000 (ablation) | Learngene individual model · ImageNet100 5-way 20-shot novel classes | accuracy: 60.86±4.48 -> 59.76±5.41 | Hurt |
| select the top network layers rather than front or middle layers as learngene (ablation) | Learngene individual model · network-position comparison | accuracy: 2% accuracy higher | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepEMD — DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover's Distance and Structured Classifiers (CVPR 2020) · better on 3 of 3 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
- Cluster-Learngene: Inheriting Adaptive Clusters for Vision Transformers (NeurIPS 2024 · similarity 0.85)
- Inheriting Generalized Learngene for Efficient Knowledge Transfer across Multiple Tasks (AAAI 2025 · similarity 0.84)
- Transformer as Linear Expansion of Learngene (AAAI 2024 · similarity 0.84)
- Vision Transformers as Probabilistic Expansion from Learngene (ICML 2024 · similarity 0.84)
- Initializing Variable-sized Vision Transformers from Learngene with Learnable Transformation (NeurIPS 2024 · similarity 0.83)
- What Do You See in Common? Learning Hierarchical Prototypes over Tree-of-Life to Discover Evolutionary Traits (ICLR 2025 · similarity 0.83)
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.
