Research map / Label-efficient and robust learning / Knowledge distillation
Spatial Ensemble: a Novel Model Smoothing Mechanism for Student-Teacher Framework
Tengteng Huang, Yifan Sun, Xun Wang, Haotian Yao, Chi Zhang
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 spatial ensemble as the model smoothing instead of no smoothing (ablation) | MoCo and BYOL · ImageNet linear eval | top-1 accuracy: fail -> 66.7 (MoCo), fail -> 67.8 (BYOL) | Improved |
| use spatial-temporal smoothing instead of temporal moving average | MoCo · ImageNet linear eval | top-1 accuracy: 67.5 -> 67.8 | Improved |
| use spatial-temporal smoothing instead of temporal moving average | BYOL · ImageNet linear eval | top-1 accuracy: 71.8 -> 72.7 | Improved |
| use neuron-wise instead of layer-wise spatial ensemble (ablation) | SE only (BYOL) · ImageNet linear eval | top-1 accuracy: 67.8 -> 71.6 | Improved |
| use neuron-wise instead of layer-wise granularity inside STS (ablation) | STS (BYOL) · ImageNet linear eval | top-1 accuracy: 72.7 -> 72.8 | No clear effect |
| use neuron-wise instead of layer-wise granularity (ablation) | STS (BYOL) · ImageNet linear eval | relative training time: 0.99x -> 1.82x | Hurt |
| use STS instead of TMA under ImageNet-C corruptions | BYOL · ImageNet-C | mean top-1 accuracy: around 1.5% | Improved |
| add STS to the FixMatch teacher | FixMatch · CIFAR-10 20 labels | top-1 accuracy: 86.68 -> 93.31 | Improved |
| add STS to the FixMatch teacher | FixMatch · CIFAR-10 40 labels | top-1 accuracy: 92.06 -> 93.89 | Improved |
| add STS to the FixMatch teacher | FixMatch · CIFAR-10 250 labels | top-1 accuracy: 94.89 -> 95.10 | Improved |
| add STS to the FixMatch teacher | FixMatch · CIFAR-10 4000 labels | top-1 accuracy: 95.77 -> 95.94 | Improved |
| add STS to the FixMatch teacher | FixMatch · CIFAR-100 200 labels | top-1 accuracy: 42.78 -> 45.85 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- How to Trade Off the Quantity and Capacity of Teacher Ensemble: Learning Categorical Distribution to Stochastically Employ a Teacher for Distillation (AAAI 2024 · similarity 0.85)
- Stable Mean Teacher for Semi-supervised Video Action Detection (AAAI 2025 · similarity 0.83)
- Less or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation (ICLR 2024 · similarity 0.83)
- See Further When Clear: Curriculum Consistency Model (CVPR 2025 · similarity 0.82)
- Improving self-training under distribution shifts via anchored confidence with theoretical guarantees (NeurIPS 2024 · similarity 0.82)
- Asymmetric Decision-Making in Online Knowledge Distillation: Unifying Consensus and Divergence (ICML 2025 · similarity 0.82)
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.
