Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Bi-Directional Distribution Alignment for Transductive Zero-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 |
|---|---|---|---|
| Replace a uniform unseen-class prior with cluster prior estimation (ablation) | Bi-VAEGAN · AWA1, AWA2, CUB, SUN | ACC^u: AWA1: 90.9 -> 91.5; AWA2: 83.1 -> 85.6; CUB: 72.5 -> 74.0; SUN: 68.4 -> 71.3 | Helped |
| Estimate the unseen prior using CPE rather than assuming a uniform prior (ablation) | Bi-VAEGAN · AWA1, AWA2, CUB, SUN | ACC^u: AWA1: 91.5 vs. 66.3; AWA2: 85.6 vs. 60.3; CUB: 74.0 vs. 76.8; SUN: 71.3 vs. 74.2 | Mixed |
| Add an inductive regressor trained on paired seen examples (ablation) | Bi-VAEGAN · AWA2, CUB, SUN | T1 accuracy: AWA2: 91.6 -> 92.3; CUB: 72.1 -> 74.3; SUN: 69.8 -> 70.8 | Helped |
| Add an adversarially trained transductive regressor (ablation) | Bi-VAEGAN · AWA2, CUB, SUN | T1 accuracy: AWA2: 91.6 -> 95.5; CUB: 72.1 -> 75.8; SUN: 69.8 -> 72.2 | Helped |
| Add L2 feature normalization to the model with TR (ablation) | Bi-VAEGAN · AWA2, CUB, SUN | T1 accuracy: AWA2: 95.5 -> 95.8; CUB: 75.8 -> 76.8; SUN: 72.2 -> 74.2 | Helped |
| Add the feature pre-tuning network (ablation) | Bi-VAEGAN · CUB (AK1) | T1 accuracy: +1.2 | Helped |
| Add the feature pre-tuning network (ablation) | Bi-VAEGAN · CUB (AK2) | T1 accuracy: 82.8 -> 82.0 (-0.8) | Hurt |
| Add the feature pre-tuning network (ablation) | Bi-VAEGAN · AWA2 | T1 accuracy: 95.8 -> 90.0 (-5.8) | Hurt |
| Add the feature pre-tuning network (ablation) | Simple-GAN · AWA2 | T1 accuracy: 92.7 -> 88.9 (-3.8) | Hurt |
| Add the feature pre-tuning network (ablation) | Simple-GAN · CUB (AK2) | T1 accuracy: +0.5 | Helped |
| Use the augmented multimodal feature space rather than visual features alone (ablation) | Bi-VAEGAN · AWA2 | T1 accuracy: 94.2 -> 95.8 | Helped |
| Use the augmented multimodal feature space rather than visual features alone (ablation) | Bi-VAEGAN · CUB (AK1) | T1 accuracy: 75.0 -> 76.8 | Helped |
Closest papers by meaning
- Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition (AAAI 2024 · similarity 0.90)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · similarity 0.87)
- Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations (NeurIPS 2024 · similarity 0.86)
- GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder (ICML 2025 · similarity 0.86)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.86)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.86)
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.