Research map / Architectures and efficiency / Quantization
Focused Quantization for Sparse CNNs
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 focused compression instead of INQ (5 bits) | ResNet-50 · ImageNet | Top-1 accuracy: 74.81 -> 74.86 | Helped |
| Use focused compression instead of Coreset | ResNet-50 · ImageNet | Compression ratio: 15.80 -> 18.08 | Helped |
| Quantize weights with 5-bit FQ (ablation) | ResNet-18 · ImageNet | Top-1 accuracy: 68.94 -> 68.36 | Hurt |
| Also quantize activations to 8-bit integers (ablation) | ResNet-18 with 5-bit FQ weights · ImageNet | Top-1 accuracy: 68.36 -> 67.89 | Hurt |
| Also quantize BN parameters to 16-bit integers (ablation) | ResNet-18 with quantized weights and activations · ImageNet | Top-1 accuracy: 67.89 -> 67.95 | No clear effect |
| Use 5-bit FQ instead of 3-bit unsigned shift quantization | Convolution accelerator · Custom convolution hardware | Number of gates: 275.2 M -> 275.6 M | No clear effect |
| Add Huffman encoding after 5-bit FQ (ablation) | Convolution accelerator · Custom convolution hardware | Number of gates: 275.6 M -> 276.4 M | Hurt |
| Use the Wasserstein threshold to select shift quantization for overlapping components and recentralized quantization for separated ones (ablation) | Focused quantization · CIFAR-10 classifier | Validation accuracy | Helped |
Closest papers by meaning
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.91)
- Quantization without Tears (CVPR 2025 · similarity 0.90)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.88)
- SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression (ICLR 2024 · similarity 0.88)
- SLiM: One-shot Quantization and Sparsity with Low-rank Approximation for LLM Weight Compression (ICML 2025 · similarity 0.88)
- SqueezeLLM: Dense-and-Sparse Quantization (ICML 2024 · 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.