Research map / Architectures and efficiency / Pruning and efficient tuning
Structural Pruning via Latency-Saliency Knapsack
Read the paperCheck 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 |
|---|---|---|---|
| HALP vs EagleEye-3G at similar FLOPs | HALP · ResNet50 ImageNet | top-1 accuracy: 77.1 -> 77.5 | Helped |
| HALP vs EagleEye-1G | HALP · ResNet50 ImageNet | speedup: 2.38x -> 2.55x | Helped |
| HALP vs Taylor-55% on ResNet101 | HALP · ResNet101 ImageNet | top-1 accuracy: 76.0 -> 77.2 | Helped |
| latency-aware neuron grouping vs fixed group sizes (ablation) | HALP · ResNet50 ImageNet | top-1 accuracy | Helped |
| very small group size (2) or very large (256) instead of latency-aware (ablation) | HALP · ResNet50 ImageNet | top-1 accuracy | Hurt |
| single-shot pruning (k=1) instead of iterative (ablation) | HALP · ResNet50 ImageNet | top-1 accuracy | Hurt |
| pruning steps k from 10 to 40 (ablation) | HALP · ResNet50 ImageNet | top-1 accuracy | No clear effect |
| knowledge distillation from RegNetY-16GF teacher (ablation) | HALP · ResNet50 ImageNet | top-1 accuracy | Helped |
| INT8 quantization on RTX3080 | HALP · ResNet50 | speedup: 9.14x (0.13%) ; 14.12x (2.38%) | Mixed |
Closest papers by meaning
- MDP: Multidimensional Vision Model Pruning with Latency Constraint (CVPR 2025 · similarity 0.91)
- When To Prune? A Policy Towards Early Structural Pruning (CVPR 2022 · similarity 0.90)
- OSSCAR: One-Shot Structured Pruning in Vision and Language Models with Combinatorial Optimization (ICML 2024 · similarity 0.87)
- Adaptive Sharpness-Aware Pruning for Robust Sparse Networks (ICLR 2024 · similarity 0.87)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.87)
- Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation (AAAI 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.