Research map / Architectures and efficiency / Quantization
Weight Fixing Networks
Christopher Subia-Waud, Srinandan Dasmahapatra
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| WFN instead of APoT 3-bit (unique weights) | ResNet-18 · ImageNet | unique parameter count: 9237 -> 164 | Improved |
| WFN delta=0.01 instead of APoT (weight entropy) | ResNet-18 · ImageNet | weight-space entropy: 5.77 -> 3.01 | Improved |
| WFN delta=0.01 instead of APoT (weight entropy) | ResNet-34 · ImageNet | weight-space entropy: 6.77 -> 3.48 | Improved |
| WFN delta=0.0075 vs APoT top-1 | ResNet-18 · ImageNet | top-1 accuracy: 69.9 -> 70.3 | Improved |
| WFN delta=0.0075 vs baseline | GoogLeNet · ImageNet | top-1 accuracy: 69.7 -> 70.9 | Improved |
| larger delta (0.0075 -> 0.015) in WFN (ablation) | ResNet-18 · ImageNet | top-1 accuracy: 70.3 -> 67.3 | Hurt |
| larger delta (0.0075 -> 0.015) in WFN (ablation) | ResNet-34 · ImageNet | top-1 accuracy: 73.0 -> 72.2 | Hurt |
| larger delta (0.0075 -> 0.015) in WFN (ablation) | ResNet-18 · ImageNet | unique parameter count: 193 -> 90 | Improved |
| WFN delta=0.015 vs APoT compression | ResNet-18 · ImageNet | compression ratio: 10.2 -> 13.4 | Improved |
| WFN delta=0.0075 vs LQ-Net | ResNet-50 · ImageNet | top-1 accuracy: 74.2 -> 76.0 | Improved |
| WFN delta=0.015 vs baseline | ResNet-34 · ImageNet | top-1 accuracy: 73.3 -> 72.2 | Hurt |
| relative-distance noise instead of absolute noise to weights (ablation) | pre-trained ResNet-18 · ImageNet noise test | accuracy tolerance | Improved |
Closest papers by meaning
- Loss-aware Weight Quantization of Deep Networks (ICLR 2018 · similarity 0.90)
- Weightless: Lossy weight encoding for deep neural network compression (ICML 2018 · similarity 0.89)
- Towards the Limit of Network Quantization (ICLR 2017 · similarity 0.89)
- Weighted-Entropy-Based Quantization for Deep Neural Networks (CVPR 2017 · similarity 0.89)
- Minimal Random Code Learning: Getting Bits Back from Compressed Model Parameters (ICLR 2019 · similarity 0.88)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · 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.
