Research map / Architectures and efficiency / Quantization
Permute, Quantize, and Fine-Tune: Efficient Compression of Neural Networks
Check 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 |
|---|---|---|---|
| Use PQF instead of BGD | Semi-supervised ResNet-50 · ImageNet | top-1 accuracy: 76.12 -> 77.15 | Helped |
| Use PQF instead of BGD | Mask R-CNN R-50 FPN · MS COCO 2017 | bounding-box AP: 33.9 -> 36.3 | Helped |
| Use PQF instead of BGD | Mask R-CNN R-50 FPN · MS COCO 2017 | mask AP: 30.8 -> 33.5 | Helped |
| Add permutation optimization (ablation) | PQF with plain k-means and SGD fine-tuning · ImageNet, ResNet-18 large blocks | accuracy: 62.29 -> 62.55 | Helped |
| Add annealed k-means (SR-C) (ablation) | PQF with permutation optimization · ImageNet, ResNet-18 large blocks | accuracy: 62.55 -> 62.92 | Helped |
| Use Adam with cosine annealing instead of plain SGD (ablation) | PQF with permutation optimization and SR-C · ImageNet, ResNet-18 large blocks | accuracy: 62.92 -> 63.31 | Helped |
| Add SR-C while leaving permutation optimization out (ablation) | PQF on Mask R-CNN R-50 FPN · MS COCO 2017 | bounding-box AP: 35.6 -> 35.8 | Helped |
| Add SR-C while leaving permutation optimization out (ablation) | PQF on Mask R-CNN R-50 FPN · MS COCO 2017 | mask AP: 33.0 -> 33.1 | Helped |
| Add permutation optimization (ablation) | PQF with SR-C · MS COCO 2017 | bounding-box AP: 35.8 -> 36.3 | Helped |
| Add permutation optimization (ablation) | PQF with SR-C · MS COCO 2017 | mask AP: 33.1 -> 33.5 | Helped |
Closest papers by meaning
- And the Bit Goes Down: Revisiting the Quantization of Neural Networks (ICLR 2020 · similarity 0.91)
- Focused Quantization for Sparse CNNs (NeurIPS 2019 · similarity 0.90)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.89)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.88)
- Qinco2: Vector Compression and Search with Improved Implicit Neural Codebooks (ICLR 2025 · similarity 0.88)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 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.