Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
DDSF: Robust Few-Shot Learning via Disentangled Subspaces with Determinantal Point Process
Xulun Ye, Yifan Mei, Kun Zhou, Zelei Wu, Jieyu Zhao
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 diversity-only anomaly filter pk (ablation) | DDSF baseline · OOD-polluted Meta-Dataset (30% noise) | accuracy: 77.1 -> 79.4 | Improved |
| add cohesion-only anomaly filter qk (ablation) | DDSF baseline · OOD-polluted Meta-Dataset (30% noise) | accuracy: 77.1 -> 79.5 | Improved |
| fuse pk and qk into score sk (ablation) | DDSF baseline · OOD-polluted Meta-Dataset (30% noise) | accuracy: 77.1 -> 80.7 | Improved |
| add DPP-guided diffusion repair of flagged outliers (ablation) | DDSF with filter sk · OOD-polluted Meta-Dataset (30% noise) | accuracy: 80.7 -> 83.4 | Improved |
| replace orthogonal loss with DPP unique volume loss Lun (ablation) | DDSF with filter and repair · OOD-polluted Meta-Dataset (30% noise) | accuracy: 83.4 -> 85.3 | Improved |
| replace orthogonal loss with DPP unique volume loss Lun (ablation) | DDSF with filter and repair · OOD-polluted Meta-Dataset (0% noise) | accuracy: 89.9 -> 91.9 | Improved |
| add foreground masking via U-2-Net (ablation) | DDSF · Meta-Dataset 0% / 30% OOD / 30% SLS | accuracy: 91.9 -> 92.1; 85.3 -> 85.2; 83.2 -> 83.5 | No clear effect |
| raise anomaly threshold tau from 0.3 to 0.6 (ablation) | DDSF · 30% OOD noise | accuracy: 85.1 -> 80.8 | Hurt |
| lower tau to 0.1 vs 0.3 (ablation) | DDSF · 30% OOD noise | accuracy: 84.5 vs 85.1 | No clear effect |
| vary loss weights lambda_sh and lambda_un (ablation) | DDSF · Meta-Dataset 0% noise | accuracy: lowest 88.11 | No clear effect |
| use DDSF instead of DETA++ at 70% OOD noise | SwinT backbone · OOD-polluted Meta-Dataset (70% noise) | accuracy: 47.0 -> 61.6 | Improved |
| use DDSF instead of DETA++ under symmetric label swapping noise | SwinT backbone · 30% SLS noise | accuracy: 75.1 -> 83.2 | Improved |
Closest papers by meaning
- EASE: Unsupervised Discriminant Subspace Learning for Transductive Few-Shot Learning (CVPR 2022 · similarity 0.86)
- Deep Learning of Determinantal Point Processes via Proper Spectral Sub-gradient (ICLR 2020 · similarity 0.86)
- State-of-art minibatches via novel DPP kernels: discretization, wavelets, and rough objectives (NeurIPS 2026 · similarity 0.86)
- Accelerating Convergence in Bayesian Few-Shot Classification (ICML 2024 · similarity 0.84)
- Robust Representation Consistency Model via Contrastive Denoising (ICLR 2025 · similarity 0.84)
- 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.
