Research map / Label-efficient and robust learning / Knowledge distillation
Variational Information Distillation for Knowledge Transfer
Check 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 |
|---|---|---|---|
| Use VID-I intermediate-layer information transfer instead of the strongest listed non-VID method | WRN-16-1 student · CIFAR-10, 100 samples/class | test accuracy: 73.40 -> 81.59 | Helped |
| Combine KD with VID-I | WRN-16-1 student · CIFAR-10, 100 samples/class | test accuracy: 76.29 -> 78.48 | Helped |
| Use VID-I instead of KD | WRN students · CIFAR-100, varying student size | test accuracy: 75.80 -> 75.25; 72.87 -> 73.31; 70.99 -> 71.51; 66.03 -> 66.32 | Mixed |
| Combine KD with VID-I | WRN-40-2 teacher and WRN students · CIFAR-100, varying student size | test accuracy: 75.25 -> 76.11 | Helped |
| Use VID-I for intermediate-layer transfer | ResNet-18 student · MIT-67, 10 samples/class | test accuracy: 47.91 -> 50.97 | Helped |
| Use VID-I for intermediate-layer transfer | ResNet-18 student · CUB-200-2011, approximately 29.95 samples/class | test accuracy: 76.69 -> 73.25 | Hurt |
| Use VID-I for transfer across heterogeneous CNN and MLP architectures | MLP student · CIFAR-10, MLP-4096 | test accuracy: 76.02 -> 85.18 | Helped |
| Use VID-I instead of the strongest listed baseline | LeNet-based student · SVHN to MNIST, 200 samples/class | test accuracy: 94.01 -> 94.17 | Helped |
| Use VID-I instead of adversarial network compression | ResNet-20 student · CIFAR-10, ResNet-164 teacher | validation accuracy: 91.92 -> 92.15 | Helped |
| Use heteroscedastic variance rather than homoscedastic variance in the variational distribution (ablation) | VID · Experiments reported in the paper | training stability and performance | Mixed |
Closest papers by meaning
- A Gift From Knowledge Distillation: Fast Optimization, Network Minimization and Transfer Learning (CVPR 2017 · similarity 0.93)
- Improved Knowledge Distillation via Teacher Assistant (AAAI 2020 · similarity 0.90)
- Progressively Knowledge Distillation via Re-parameterizing Diffusion Reverse Process (AAAI 2024 · similarity 0.90)
- CIFD: Controlled Information Flow to Enhance Knowledge Distillation (NeurIPS 2024 · similarity 0.88)
- All You Need in Knowledge Distillation Is a Tailored Coordinate System (AAAI 2025 · similarity 0.88)
- Cooperative Knowledge Distillation: A Learner Agnostic Approach (AAAI 2024 · similarity 0.87)
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.