Research map / Label-efficient and robust learning / Recognition and long-tail
Virtual Class Enhanced Discriminative Embedding 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 standard softmax with Virtual Softmax using a dynamic virtual negative class | CNN classification · CIFAR10 | error rate (%): 7.15 -> 6.68 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | CNN classification · CIFAR100 | error rate (%): 27.7 -> 26.02 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | CNN classification · MNIST | error rate (%): 0.35 -> 0.28 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | CNN classification · SVHN | error rate (%): 2.11 -> 1.93 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | GoogLeNet V1 · CUB200 | accuracy (%): 73.5 -> 77.1 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | GoogLeNet V2 · CUB200 | accuracy (%): 77.2 -> 81.1 | Helped |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | ResNet face verification model · LFW | verification accuracy (%): 99.10 -> 99.46 | Helped |
| Replace standard softmax with Virtual Softmax | ResNet face verification model · LFW | verification accuracy (%): 99.57 -> 99.46 | Hurt |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | ResNet face verification model · SLLFW | verification accuracy (%): 94.59 -> 95.85 | Helped |
| Replace standard softmax with Virtual Softmax | ResNet face verification model · SLLFW | verification accuracy (%): 96.45 -> 95.85 | Hurt |
| Replace standard softmax with Virtual Softmax using a dynamic virtual negative class | CNN classification · ImageNet32 | Top1 accuracy (%): 47.63 -> 48.84 | Helped |
| Increase network width from t=1 to t=4 (ablation) | Virtual Softmax CNN · CIFAR10 | error rate (%): 8.95 -> 6.68 | Helped |
Closest papers by meaning
- Softmax Dissection: Towards Understanding Intra- and Inter-Class Objective for Embedding Learning (AAAI 2020 · similarity 0.89)
- Learning Towards The Largest Margins (ICLR 2022 · similarity 0.88)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.86)
- Large-Margin Softmax Loss for Convolutional Neural Networks (ICML 2016 · similarity 0.86)
- Training Deep Neural Networks with Virtual Smoothing Classes (AAAI 2025 · similarity 0.83)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.83)
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.