AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Episode-Based Prototype Generating Network for Zero-Shot Learning

Yunlong Yu, Zhong Ji, Jungong Han, Zhongfei Zhang

CVPR 2020 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Generative feature synthesis for zero-shot · 181 citations (OpenAlex)

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 changedWhereMeasuredEffect
Train with the episode-based paradigm instead of batch-based PGN training (ablation)PGN · AwA1, AwA2, CUB, FLOH: 65.3 -> 71.2; 61.0 -> 64.6; 52.5 -> 56.2; 70.0 -> 76.5Improved
Use 5 rather than 10 mimetic unseen classes per episode (ablation)E-PGN · AwA1, AwA2, CUB, FLOH: 68.0 vs 71.2; 63.3 vs 64.6; 54.4 vs 56.2; 73.2 vs 76.5Hurt
Use 15 rather than 10 mimetic unseen classes per episode (ablation)E-PGN · AwA1, AwA2, CUB, FLOH: 71.1 vs 71.2; 63.1 vs 64.6; 54.3 vs 56.2; 75.2 vs 76.5Mixed
Add MCE loss (ablation)E-PGN · AwA1, AwA2, CUB, FLOH: 66.8 -> 71.2; 56.1 -> 64.6; 48.8 -> 56.2; 74.4 -> 76.5Improved
Add visual-to-semantic regression loss (ablation)E-PGN · AwA1, AwA2, CUB, FLOH: 69.6 -> 71.2; 63.0 -> 64.6; 54.7 -> 56.2; 74.1 -> 76.5Improved
Add semantic-to-visual regression loss (ablation)E-PGN · AwA1, AwA2, CUB, FLOH: 70.4 -> 71.2; 64.3 -> 64.6; 50.4 -> 56.2; 75.2 -> 76.5Improved
Use MCE rather than traditional cross-entropy (ablation)E-PGN · AwA1, AwA2, CUB, FLOclassification performanceImproved
Use Euclidean rather than cosine distance (ablation)E-PGN · four ZSL datasetsclassification performanceImproved
Use E-PGN rather than the strongest listed baselineZero-shot classification · AwA1, AwA2, CUB, FLOT: 71.1 -> 74.4; 70.4 -> 73.4; 71.1 -> 72.4; 83.9 -> 85.7Improved
Use E-PGN rather than the strongest listed baselineGeneralized zero-shot classification · AwA1, AwA2, CUB, FLOH: 64.1 -> 71.2; 63.9 -> 64.6; 54.3 -> 56.2; 68.3 -> 76.5Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy