Research map / Architectures and efficiency / RNNs and normalization
A Greedy Algorithm for Quantizing Neural Networks
Eric Lybrand, Rayan Saab
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 |
|---|---|---|---|
| quantize with GPFQ instead of memoryless scalar quantization (MSQ) | CNN, 2-bit alphabet · CIFAR10 test (best C_alpha) | top-1 accuracy: 0.2800 -> 0.8036 | Improved |
| quantize with GPFQ instead of MSQ | CNN, 4-bit alphabet · CIFAR10 test | top-1 accuracy: 0.8443 -> 0.8888 | Improved |
| quantize with GPFQ instead of MSQ | VGG16 fully connected layers · ILSVRC2012 top-1 | accuracy: 0.69485 -> 0.70075 | Improved |
| quantize with GPFQ instead of MSQ at large alphabet scalar | VGG16 · ILSVRC2012 top-1 | accuracy: 0.53855 -> 0.68335 | Improved |
| increase alphabet scalar C_alpha from 2 to 6 for GPFQ (ablation) | CNN 2-bit · CIFAR10 test | top-1 accuracy: 0.7522 -> 0.5365 | Hurt |
| increase bit budget from log2(3) to 4 bits for GPFQ (ablation) | CNN · CIFAR10 test | top-1 accuracy: 0.7487 -> 0.8888 | Improved |
| increase bit budget from log2(3) to 4 bits for MSQ (ablation) | CNN · CIFAR10 test | top-1 accuracy: 0.1464 -> 0.8443 | Improved |
| increase alphabet scalar C_alpha from 3 to 5 for MSQ (ablation) | VGG16 · ILSVRC2012 top-1 | accuracy: 0.69485 -> 0.53855 | Hurt |
| increase alphabet scalar C_alpha from 3 to 5 for GPFQ (ablation) | VGG16 · ILSVRC2012 top-1 | accuracy: 0.70075 -> 0.68335 | Hurt |
Closest papers by meaning
- Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data (AAAI 2019 · similarity 0.89)
- LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks (ECCV 2018 · similarity 0.88)
- Proximal Mean-Field for Neural Network Quantization (ICCV 2019 · similarity 0.88)
- Variational Network Quantization (ICLR 2018 · similarity 0.88)
- Network Quantization With Element-Wise Gradient Scaling (CVPR 2021 · similarity 0.88)
- Quantization without Tears (CVPR 2025 · 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.
