Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network
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 |
|---|---|---|---|
| select task-relevant query descriptors using DMNN rather than accumulating all descriptors | DN4/NBNN · miniImageNet 1-shot, Conv-4 | accuracy: 54.66 -> 55.77 | Helped |
| select task-relevant query descriptors using MNN rather than accumulating all descriptors (ablation) | DN4/NBNN · CUB 1-shot, Conv-4 | accuracy: 73.42 -> 78.10 | Helped |
| use DMNN's relative similarity rather than MNN's absolute similarity for selection (ablation) | MN4 · CUB 1-shot, Conv-4 | accuracy: 78.10 -> 78.36 | Helped |
| use DMNN's relative similarity rather than MNN's absolute similarity for selection (ablation) | MN4 · tieredImageNet 1-shot, Conv-4 | accuracy: 57.01 -> 56.99 | No clear effect |
| use DMNN's relative similarity rather than MNN's absolute similarity for selection (ablation) | MN4 · miniImageNet 1-shot, ResNet-12 | accuracy: 66.53 -> 66.58 | No clear effect |
| replace MNN selection with top-k online discriminative mining (ablation) | DN4 · CUB 1-shot, Conv-4 | accuracy: 78.10 vs 77.02 | Hurt |
| replace cosine similarity scores with class-vote counts (ablation) | DMN4 · miniImageNet 1-shot, Conv-4 | accuracy: 55.77 -> 53.97 | Hurt |
| replace cosine similarity scores with class-vote counts (ablation) | DMN4 · miniImageNet 5-shot, Conv-4 | accuracy: 74.22 -> 72.60 | Hurt |
| use deeper ResNet-12 descriptors with a larger receptive field, where descriptor filtering offers less gain (ablation) | MN4 versus DN4 · miniImageNet 1-shot | accuracy: 54.66 -> 55.57; 65.35 -> 66.53 | Mixed |
| add MNN-based pseudo-labeling from unlabeled examples | DN4 semi-supervised · miniImageNet 1-shot, without distractors | accuracy: 51.46 -> 53.48 | Helped |
| add MNN-based pseudo-labeling from unlabeled examples | DN4 semi-supervised · miniImageNet 5-shot, without distractors | accuracy: 68.75 -> 71.06 | Helped |
Closest papers by meaning
- Revisiting Local Descriptor Based Image-To-Class Measure for Few-Shot Learning (CVPR 2019 · similarity 0.92)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.89)
- Learning To Affiliate: Mutual Centralized Learning for Few-Shot Classification (CVPR 2022 · similarity 0.88)
- Few-Shot Classification With Feature Map Reconstruction Networks (CVPR 2021 · similarity 0.88)
- Dense Classification and Implanting for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.87)
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.