Research map / Architectures and efficiency / Pruning and efficient tuning
Beyond Filter Pruning: Top-K Spatial Selection for Efficient Neural Networks
Sarthak Ketanbhai Modi, Hans Soegeng, Thomas Peyrin
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 |
|---|---|---|---|
| prune spatial positions within kernels (TKP, K_keep=1) instead of filter pruning (DepGraph) | ResNet56 · CIFAR-10 | top-1 accuracy: 88.26 -> 91.80 | Improved |
| TKP instead of ETP | VGG-19 · CIFAR-100 | top-1 accuracy: 71.30 -> 72.77 | Improved |
| TKP on ResNet50 at 3.51x FLOP reduction | ResNet50 · ImageNet-1k | top-1 accuracy: 76.15 -> 76.49 | Improved |
| TKP on ViT-B/16 at 1.78x FLOP reduction | ViT-B/16 · ImageNet-1k | top-1 accuracy: 81.07 -> 80.93 | No clear effect |
| TKP instead of MDP for measured speed-up | ResNet56 · CIFAR-10 | speed-up: 2.97x -> 3.94x | Improved |
| TKP instead of Isomorphic pruning | MobileNetV3-Large · ImageNet-1k | top-1 accuracy: 70.8 -> 72.8 | Improved |
| combine TKP with QAT instead of DepGraph with QAT | ResNet-18 · CIFAR-100 | accuracy: 76.96% -> 81.54% | Improved |
| combine TKP with QAT instead of QAT-INT8 alone | VGG-19 · CIFAR-100 | accuracy: 72.14% -> 71.45% | Hurt |
| TKP instead of Wanda++ at 2:4 | LLaMA-2-7B · WikiText-2 | perplexity: 10.15 -> 9.31 | Improved |
| TKP at 50% sparsity vs dense | LLaMA-2-7B · WikiText-2 | perplexity: 6.00 -> 5.59 | Improved |
| TKP instead of Wanda++ at 2:4 | BLIP-Base · Flickr30k | BLEU-4: 28.93 -> 29.93 | Improved |
| add exponential regularization stage before Top-K (ablation) | TKP (CNNs) · CIFAR-10/100 (ResNet-56, VGG-8, GoogleNet, DenseNet-121) | accuracy | Improved |
Closest papers by meaning
- HRank: Filter Pruning Using High-Rank Feature Map (CVPR 2020 · similarity 0.90)
- Interspace Pruning: Using Adaptive Filter Representations To Improve Training of Sparse CNNs (CVPR 2022 · similarity 0.90)
- VLM-Pruner: Buffering for Spatial Sparsity in an Efficient VLM Centrifugal Token Pruning Paradigm (CVPR 2026 · similarity 0.88)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 2024 · similarity 0.88)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 2024 · similarity 0.87)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · 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.
