Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Zero and Few Shot Learning With Semantic Feature Synthesis and Competitive Learning
Jiechao Guan, Zhiwu Lu, Tao Xiang, Aoxue Li, An Zhao, Ji-Rong Wen
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 feature synthesis (SFSP) to bidirectional projection learning (ablation) | BPL · AwA, CUB, SUN | ZSL accuracy: 3% to 7% improvement | Improved |
| add competitive learning (ablation) | BPL · AwA, CUB, SUN | ZSL accuracy: 1-3% gains | Improved |
| use bidirectional instead of single-direction projection learning (ablation) | BPL0 · AwA, CUB, SUN | ZSL accuracy | Improved |
| replace no-ambiguity-handling with competitive learning (ablation) | BPL · AwA (pure ZSL) | accuracy: 62.9 -> 74.1 | Improved |
| replace minimum entropy with competitive learning (ablation) | BPL · AwA (pure ZSL) | accuracy: 67.1 -> 74.1 | Improved |
| replace minimum entropy with competitive learning (ablation) | BPL · CUB (pure ZSL) | accuracy: 59.0 -> 61.9 | Improved |
| replace minimum entropy with competitive learning (ablation) | BPL · SUN (pure ZSL) | accuracy: 59.6 -> 63.2 | Improved |
| use few-shot real samples (SFSP1) rather than base class samples (SFSP2) for synthesis (ablation) | BPL · ImNet FSL K=1 | hit@1: 22.7 -> 22.8 | No clear effect |
| use SFSP1 instead of SFSP2 (ablation) | BPL · ImNet FSL | training runtime (min): 2.3 -> 1.2 | Improved |
| add base class knowledge transfer via joint projection learning (ablation) | BPL (novel) · ImNet FSL | hit accuracy | Improved |
| use stronger WRN features instead of Simple CNN | Ours · mini-ImageNet 5-way 5-shot | accuracy: 65.28 -> 75.62 | Improved |
| use proposed model instead of PPA | BPL · mini-ImageNet 5-way 5-shot (WRN) | accuracy: 73.74 -> 75.62 | Improved |
Closest papers by meaning
- Transferrable Feature and Projection Learning with Class Hierarchy for Zero-Shot Learning (IJCV 2020 · similarity 0.92)
- Domain-Invariant Projection Learning for Zero-Shot Recognition (NeurIPS 2018 · similarity 0.91)
- Zero-Shot Learning Using Synthesised Unseen Visual Data with Diffusion Regularisation (TPAMI 2017 · similarity 0.91)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.89)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.89)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · similarity 0.88)
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.
