Research map / Architectures and efficiency / Quantization
Effective Training of Convolutional Neural Networks With Low-Bitwidth Weights and Activations
Bohan Zhuang, Mingkui Tan, Jing Liu, Lingqiao Liu, Ian D. Reid, Chunhua Shen
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 |
|---|---|---|---|
| add two-stage training (TS) (ablation) | 2-bit ResNet-50 DoReFa · ImageNet val 2W2A | Top-1 accuracy (%): 70.19 -> 70.92 | Improved |
| add progressive precision (PP) (ablation) | 2-bit ResNet-50 DoReFa · ImageNet val 2W2A | Top-1 accuracy (%): 70.19 -> 70.78 | Improved |
| add TS and PP together (ablation) | 2-bit ResNet-50 DoReFa · ImageNet val 2W2A | Top-1 accuracy (%): 70.19 -> 71.13 | Improved |
| add stochastic precision (SP) | DoReFa-Net ResNet-50 · ImageNet val | Top-1 accuracy (%): 70.19 -> 72.23 | Improved |
| add stochastic precision (SP) | BiReal-Net ResNet-18 · ImageNet val | Top-1 accuracy (%): 56.43 -> 58.81 | Improved |
| add stochastic precision (SP) | GroupNet ResNet-18 · ImageNet val | Top-1 accuracy (%): 64.82 -> 65.89 | Improved |
| add random weight/activation quantization to layerdrop (ablation) | GroupNet + layerdrop · ImageNet val | Top-1 accuracy (%): 65.7 -> 65.0 | Hurt |
| add random weight/activation quantization to blockdrop (ablation) | GroupNet + blockdrop · ImageNet val | Top-1 accuracy (%): 65.6 -> 65.9 | Improved |
| add joint knowledge distillation with same-size teacher | 2-bit ResNet-50 · ImageNet val 2W2A | Top-1 accuracy (%): 70.19 -> 71.96 | Improved |
| jointly train teacher instead of freezing it (ablation) | 2-bit PreResNet-18 KD · ImageNet val 2W2A | Top-1 accuracy (%): 65.09 -> 65.67 | Improved |
| use deeper PreResNet-50 teacher instead of PreResNet-34 (ablation) | 2-bit PreResNet-18 · ImageNet val 2W2A | Top-1 accuracy (%): 65.82 -> 65.86 | No clear effect |
| add attention-transfer guidance to posterior distillation (ablation) | 2-bit ResNet-50 KD · ImageNet val 2W2A | Top-1 accuracy (%): 71.40 -> 71.96 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BiReal-Net — Bi-Real Net: Enhancing the Performance of 1-bit CNNs with Improved Representational Capability and Advanced Training Algorithm (ECCV 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Effective Low-Bitwidth Convolutional Neural Networks (CVPR 2018 · similarity 0.97)
- LEARNED STEP SIZE QUANTIZATION (ICLR 2020 · similarity 0.92)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.91)
- Post training 4-bit quantization of convolutional networks for rapid-deployment (NeurIPS 2019 · similarity 0.90)
- Not All Bits have Equal Value: Heterogeneous Precisions via Trainable Noise (NeurIPS 2022 · similarity 0.90)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · 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.
