Research map / Architectures and efficiency / Quantization
And the Bit Goes Down: Revisiting the Quantization of Neural Networks
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 |
|---|---|---|---|
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, small blocks, k=256 | top-1 accuracy: 64.76 -> 65.81 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, small blocks, k=512 | top-1 accuracy: 66.31 -> 67.15 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, small blocks, k=1024 | top-1 accuracy: 67.28 -> 67.87 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, small blocks, k=2048 | top-1 accuracy: 67.88 -> 68.26 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, large blocks, k=256 | top-1 accuracy: 60.46 -> 61.18 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, large blocks, k=512 | top-1 accuracy: 63.21 -> 63.99 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, large blocks, k=1024 | top-1 accuracy: 64.74 -> 65.72 | Helped |
| Use activation-aware quantization instead of standard PQ (ablation) | ResNet-18 with distillation fine-tuning · ImageNet, large blocks, k=2048 | top-1 accuracy: 65.94 -> 66.50 | Helped |
| Use distillation instead of image-label fine-tuning (ablation) | Activation-aware ResNet-18 quantization · ImageNet, small blocks, k=256 | top-1 accuracy: 65.55 -> 65.81 | Helped |
| Use distillation instead of image-label fine-tuning (ablation) | Activation-aware ResNet-18 quantization · ImageNet, small blocks, k=512 | top-1 accuracy: 66.82 -> 67.15 | Helped |
| Use distillation instead of image-label fine-tuning (ablation) | Activation-aware ResNet-18 quantization · ImageNet, small blocks, k=1024 | top-1 accuracy: 67.53 -> 67.87 | Helped |
| Use distillation instead of image-label fine-tuning (ablation) | Activation-aware ResNet-18 quantization · ImageNet, small blocks, k=2048 | top-1 accuracy: 67.99 -> 68.26 | Helped |
Closest papers by meaning
- Focused Quantization for Sparse CNNs (NeurIPS 2019 · similarity 0.91)
- Permute, Quantize, and Fine-Tune: Efficient Compression of Neural Networks (CVPR 2021 · similarity 0.91)
- Adaptive Loss-Aware Quantization for Multi-Bit Networks (CVPR 2020 · similarity 0.90)
- Quantization without Tears (CVPR 2025 · similarity 0.90)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.90)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.90)
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.