Research map / Optimization / Optimizers
Relaxed Quantization for Discretized 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 |
|---|---|---|---|
| Train with relaxed quantization instead of full precision | LeNet-5 · MNIST | test error (%): 0.64 -> 0.55 | Helped |
| Train with relaxed quantization instead of full precision | VGG-7 · CIFAR 10 | test error (%): 6.95 -> 6.70 | Helped |
| Use 4/4-bit RQ instead of 8/8-bit RQ | VGG-7 · CIFAR 10 | test error (%): 6.70 -> 8.43 | Hurt |
| Use 2/2-bit RQ instead of 8/8-bit RQ | VGG-7 · CIFAR 10 | test error (%): 6.70 -> 11.75 | Hurt |
| Use the straight-through estimator variant rather than concrete sampling | RQ · CIFAR 10 | test error (%): 8.43 -> 7.96; 11.75 -> 9.08 | Helped |
| Use the straight-through estimator variant rather than concrete sampling | RQ · ImageNet, ResNet-18 | top-1 error (%): 30.03 -> 30.37; 31.35 -> 31.85; 34.90 -> 36.65; 38.48 -> 37.54 | Mixed |
| Train with relaxed quantization instead of full precision | ResNet-18 · ImageNet | top-1 error (%): 30.46 -> 30.03 | Helped |
| Use 5/5-bit RQ instead of 8/8-bit RQ | MobileNet · ImageNet | top-1 error (%): 29.57 -> 38.62 | Hurt |
| Use 8/8-bit SR+DR instead of 8/8-bit RQ | MobileNet · ImageNet | top-1 error (%): 29.57 -> 28.70 | Helped |
Closest papers by meaning
- Up or Down? Adaptive Rounding for Post-Training Quantization (ICML 2020 · similarity 0.89)
- Network Quantization With Element-Wise Gradient Scaling (CVPR 2021 · similarity 0.88)
- Variational Network Quantization (ICLR 2018 · similarity 0.88)
- Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data (AAAI 2019 · similarity 0.88)
- LOTION: Smoothing the Optimization Landscape for Quantized Training (NeurIPS 2026 · similarity 0.86)
- Layer-wise Quantization for Quantized Optimistic Dual Averaging (ICML 2025 · similarity 0.86)
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.