Research map / Architectures and efficiency / Pruning and efficient tuning
Pruning vs Quantization: Which is Better?
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 |
|---|---|---|---|
| quantize to 4 bits instead of 75% magnitude pruning | fine-tuned networks · MobileNet-V2 ImageNet | accuracy (%): 48.0 -> 70.9 | Helped |
| quantize to 4 bits instead of 75% pruning | fine-tuned networks · OPT-350m WikiText-103 | perplexity: 35.4 -> 15.3 | Helped |
| quantize to 4 bits instead of pruning | fine-tuned networks · EfficientDet COCO | mAP: 24.2 -> 37.8 | Helped |
| quantize to 4 bits instead of pruning | fine-tuned networks · ResNet-18 ImageNet | accuracy (%): 69.3 -> 70.0 | Helped |
| prune to 87.5% sparsity instead of 2-bit quantization | fine-tuned ResNet-50 · ImageNet | accuracy (%): 72.3 -> 74.3 | Helped |
| prune instead of quantize across all models and bit-widths | fine-tuned networks · 9 models, 4 tasks | accuracy | Mixed |
| double the fine-tuning length for pruning (ablation) | magnitude pruning ResNet-50 · ImageNet 2-bit | accuracy (%): 74.3 -> 74.8 | Helped |
| double the fine-tuning length for pruning (ablation) | magnitude pruning ViT · ImageNet 2-bit | accuracy (%): 69.4 -> 71.6 | Helped |
| use quantization instead of pruning on Gaussian weights | weight tensors · standard normal distribution | SNR (dB): 5.6 -> 19.1 | Helped |
| use pruning instead of quantization for high-kurtosis tensors at very high compression | weight tensors · real model-zoo tensors | SNR | Mixed |
Closest papers by meaning
- Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (CVPR 2025 · similarity 0.89)
- Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruning (NeurIPS 2022 · similarity 0.89)
- Automatic Neural Network Compression by Sparsity-Quantization Joint Learning: A Constrained Optimization-Based Approach (CVPR 2020 · similarity 0.89)
- OPQ: Compressing Deep Neural Networks with One-shot Pruning-Quantization (AAAI 2021 · similarity 0.88)
- What Makes a Good Prune? Maximal Unstructured Pruning for Maximal Cosine Similarity (ICLR 2024 · similarity 0.88)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.88)
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.