Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Information Maximization for Few-Shot 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 label-marginal entropy to CE + conditional entropy (W only) (ablation) | TIM-GD · mini-ImageNet 1-shot 5-way | accuracy (%): 35.3 -> 73.9 | Helped |
| add label-marginal entropy to CE + conditional entropy (W only) (ablation) | TIM-ADM · mini-ImageNet 1-shot 5-way | accuracy (%): 36.0 -> 73.6 | Helped |
| add conditional entropy to CE + marginal entropy (W only) (ablation) | TIM-GD · mini-ImageNet 1-shot 5-way | accuracy (%): 66.1 -> 73.9 | Helped |
| add mutual information terms to CE only (W only) (ablation) | TIM-GD · mini-ImageNet 1-shot 5-way | accuracy (%): 60.7 -> 73.9 | Helped |
| fine-tune whole network {phi,W} instead of only W (ablation) | TIM-GD · mini-ImageNet 1-shot 5-way | accuracy (%): 73.9 -> 67.2 | Hurt |
| fine-tune whole network {phi,W} instead of only W (ablation) | TIM-GD · mini-ImageNet 5-shot 5-way | accuracy (%): 85.0 -> 84.7 | Hurt |
| remove marginal entropy when fine-tuning whole network (ablation) | TIM-GD {phi,W} · mini-ImageNet 1-shot 5-way | accuracy (%): 67.2 -> 62.7 | Mixed |
| use ADM solver instead of gradient descent | TIM · mini-ImageNet 5-shot 5-way | inference time (s): 2.2 × 10+0 -> 1.2 × 10−1 | Helped |
| use ADM solver instead of gradient descent | TIM · mini-ImageNet 1-shot 5-way | accuracy (%): 73.9 -> 73.6 | No clear effect |
| replace entropy fine-tuning of whole network (Ent-min) with TIM on W only | TIM-GD · mini-ImageNet 5-shot 5-way | inference time (s): 2.1 × 10+1 -> 2.2 × 10+0 | Helped |
| TIM vs best prior method under domain shift | TIM-GD · mini-ImageNet to CUB 5-shot | accuracy (%): 67.0 -> 71.0 | Helped |
Closest papers by meaning
- TIM++: Transductive Information Maximization for Few-Shot CLIP (AAAI 2026 · similarity 0.91)
- Laplacian Regularized Few-Shot Learning (ICML 2020 · similarity 0.90)
- Realistic evaluation of transductive few-shot learning (NeurIPS 2021 · similarity 0.90)
- Transductive Zero-Shot and Few-Shot CLIP (CVPR 2024 · similarity 0.87)
- UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning (CVPR 2025 · similarity 0.86)
- BECLR: Batch Enhanced Contrastive Few-Shot Learning (ICLR 2024 · similarity 0.84)
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.