Research map / Label-efficient and robust learning / Knowledge distillation
Differentiable Feature Aggregation Search for Knowledge Distillation
Yushuo Guan, Pengyu Zhao, Bingxuan Wang, Yuanxing Zhang, Cong Yao, Kaigui Bian, Jian Tang
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 |
|---|---|---|---|
| use searched feature aggregation instead of KD output distillation | WRN_16_4 student · CIFAR-100 | accuracy (%): 78.4 -> 79.74 | Improved |
| use searched aggregation instead of last-layer-only features (Last) (ablation) | WRN_16_4 student · CIFAR-100 | accuracy (%): 79.11 -> 79.74 | Improved |
| use searched aggregation instead of last-layer-only features (Last) (ablation) | WRN_28_2 student · CIFAR-100 | accuracy (%): 77.86 -> 78.18 | Improved |
| use searched aggregation instead of last-layer-only features (Last) (ablation) | WRN_16_2 student · CIFAR-100 | accuracy (%): 75.51 -> 75.85 | Improved |
| use searched aggregation instead of random weights (ablation) | WRN_16_4 student · CIFAR-100 | accuracy (%): 78.92 -> 79.74 | Improved |
| use searched aggregation instead of random weights (ablation) | WRN_16_2 student · CIFAR-100 | accuracy (%): 74.25 -> 75.85 | Improved |
| use searched aggregation instead of average weights (ablation) | WRN_16_4 student · CIFAR-100 | accuracy (%): 78.81 -> 79.74 | Improved |
| use searched aggregation instead of average weights (ablation) | WRN_28_2 student · CIFAR-100 | accuracy (%): 76.99 -> 78.18 | Improved |
| add DFA feature distillation over SP | WRN_28_2 student · CIFAR-100 | accuracy (%): 77.41 -> 78.18 | Improved |
| search aggregation weights on CIFAR-100 instead of CINIC-10 (DFA-T) (ablation) | DFA, ShuffleNetV2 x=1.0 -> x=0.5 · CINIC-10 | accuracy (%): 79.97 -> 79.38 | Hurt |
| search aggregation weights on CIFAR-100 instead of CINIC-10 (DFA-T) (ablation) | DFA, ShuffleNetV2 x=2.0 -> x=1.0 · CINIC-10 | accuracy (%): 85.38 -> 85.41 | No clear effect |
| search aggregation weights on CIFAR-100 instead of CINIC-10 (DFA-T) (ablation) | DFA, ShuffleNetV2 x=2.0 -> x=0.5 · CINIC-10 | accuracy (%): 79.51 -> 79.45 | No clear effect |
Closest papers by meaning
- DistPro: Searching a Fast Knowledge Distillation Process via Meta Optimization (ECCV 2022 · similarity 0.89)
- How to Trade Off the Quantity and Capacity of Teacher Ensemble: Learning Categorical Distribution to Stochastically Employ a Teacher for Distillation (AAAI 2024 · similarity 0.87)
- CIFD: Controlled Information Flow to Enhance Knowledge Distillation (NeurIPS 2024 · similarity 0.86)
- Hybrid Data-Free Knowledge Distillation (AAAI 2025 · similarity 0.86)
- Less or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation (ICLR 2024 · similarity 0.86)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · 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.
