Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Boosting Transductive Few-Shot Fine-Tuning With Margin-Based Uncertainty Weighting and Probability Regularization
Ran Tao, Hao Chen, Marios Savvides
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 margin-based uncertainty weighting (top-2 probabilities) to transductive fine-tuning (ablation) | TF · Meta-Dataset Imagenet-only evaluation | accuracy: around 6% performance gain over equally weighting unlabeled samples | Improved |
| Use top-2 probabilities rather than all probabilities for uncertainty weighting (ablation) | Margin-based uncertainty weighting · Meta-Dataset Imagenet-only evaluation | accuracy | Improved |
| Add margin-based uncertainty weighting to TF (ablation) | TF · Meta-Dataset, 10 datasets | accuracy: from 1.27% on MSCOCO to 13.54% on the Traffic sign | Improved |
| Add probability regularization to TF without MW (ablation) | TF · Meta-Dataset, 10 datasets | accuracy: over 7 datasets from 0.29% on DTD to 1.79% on Traffic sign | Improved |
| Add probability regularization to TF with MW (ablation) | TF with MW · Meta-Dataset, 10 datasets | accuracy: from 0.45% on Fungi to 4.27% on the Traffic sign | Improved |
| Use uniform prior and a per-query estimated marginal distribution for probability regularization (ablation) | Probability regularization · ILSVRC-2012 validation, uniform and stochastic query sets | accuracy | Improved |
| Use transductive fine-tuning rather than support-set-only fine-tuning (ablation) | TF-MP · Meta-Dataset, 1-shot cases | accuracy: over 10% compared with fine-tuning | Improved |
| Use transductive fine-tuning without uncertainty weighting or probability regularization (ablation) | TF · Meta-Dataset, 10 datasets | accuracy | Hurt |
| Add TF-MP to TSA | TSA · Meta-Dataset All-datasets evaluation | accuracy: 0.82% average margin | Improved |
Closest papers by meaning
- Realistic evaluation of transductive few-shot learning (NeurIPS 2021 · similarity 0.88)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.85)
- UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning (CVPR 2025 · similarity 0.85)
- LT-Soups: Bridging Head and Tail Classes via Subsampled Model Soups (NeurIPS 2025 · similarity 0.84)
- Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework (NeurIPS 2025 · similarity 0.84)
- Tuning Stable Rank Shrinkage: Aiming at the Overlooked Structural Risk in Fine-tuning (CVPR 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.
