Research map / Architectures and efficiency / Pruning and efficient tuning
Partial Forward Blocking: A Novel Data Pruning Paradigm for Lossless Training Acceleration
Dongyue Wu, Zilin Guo, Jialong Zuo, Nong Sang, Changxin Gao
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 adaptive centroid weight w_j in KDE (ablation) | PFB · CIFAR-100 (70% pruned) | top-1 accuracy: 77.4 -> 77.9 | Improved |
| add adaptive centroid weight w_j in KDE (ablation) | PFB · CIFAR-100 (30% pruned) | top-1 accuracy: 78.9 -> 79.1 | Improved |
| add adaptive centroid weight w_j in KDE (ablation) | PFB · CIFAR-100 (50% pruned) | top-1 accuracy: 78.4 -> 78.8 | Improved |
| use Silverman bandwidth instead of identity matrix (ablation) | PFB · CIFAR-100 (70% pruned) | top-1 accuracy: 75.1 -> 77.9 | Improved |
| use Silverman bandwidth instead of identity matrix (ablation) | PFB · CIFAR-100 (30% pruned) | top-1 accuracy: 77.4 -> 79.1 | Improved |
| use Silverman bandwidth instead of identity matrix (ablation) | PFB · CIFAR-100 (50% pruned) | top-1 accuracy: 76.5 -> 78.8 | Improved |
| use Silverman instead of Scott bandwidth rule (ablation) | PFB · CIFAR-100 (30% pruned) | top-1 accuracy: 79.0 -> 79.1 | No clear effect |
| increase random-term upper bound b to 10 (ablation) | PFB · CIFAR-100 | top-1 accuracy | Hurt |
| vary random-term upper bound b within [0.001, 1] (ablation) | PFB · CIFAR-100 | top-1 accuracy | No clear effect |
| apply blocking at stem layers (stage-0) instead of stage-1 (ablation) | PFB · ImageNet-1k | top-1 accuracy | Hurt |
| vary number of centroids N_C and feature dimension D within normal range (ablation) | PFB · CIFAR-100 | top-1 accuracy | No clear effect |
| vary batch size (ablation) | PFB · ImageNet-1k | top-1 accuracy | No clear effect |
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.
- InfoBatch* — InfoBatch: Lossless Training Speed Up by Unbiased Dynamic Data Pruning (ICLR 2024) · better on 3 of 3 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
- OrderDP: A Theoretically Guaranteed Lossless Dynamic Data Pruning Framework (ICLR 2026 · similarity 0.89)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.88)
- Exploring Learning Complexity for Efficient Downstream Dataset Pruning (ICLR 2025 · similarity 0.88)
- Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration (ICML 2026 · similarity 0.87)
- InfoBatch: Lossless Training Speed Up by Unbiased Dynamic Data Pruning (ICLR 2024 · similarity 0.87)
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 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.
