Research map / Architectures and efficiency / Quantization
NIPQ: Noise Proxy-Based Integrated Pseudo-Quantization
Juncheol Shin, Junhyuk So, Sein Park, Seungyeop Kang, Sungjoo Yoo, Eunhyeok Park
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 |
|---|---|---|---|
| Use NIPQ mixed precision instead of LSQ | MobileNet-v2 · ImageNet | Top-1 accuracy (%): 70.46 -> 71.52 | Improved |
| Use NIPQ BOP-constrained mixed precision instead of LSQ | MobileNet-v2 · ImageNet | Top-1 accuracy (%): 70.46 -> 72.26 | Improved |
| Add knowledge distillation | NIPQ mixed-precision quantization · ImageNet, MobileNet-v2 | Top-1 accuracy (%): 71.52 -> 72.16 | Improved |
| Add truncation rather than using min-max quantization | Pseudo-quantization weight training · GLUE, BERT-base, weight quantization | score: At 3-bit, MNLI: 31.82 vs 81.88; CoLA: 0.02 vs 53.03 | Improved |
| Add stochastic rounding of bit-width during training (ablation) | NIPQ · CIFAR-100, MobileNet-v2 | accuracy | Improved |
| Add post-training batch-normalization statistics update (ablation) | NIPQ · CIFAR-100, MobileNet-v2, 1.0 GBlops | Top-1 accuracy: 70.45 -> 72.99 | Improved |
| Add late-stage QAT finetuning (ablation) | NIPQ · CIFAR-100, MobileNet-v2, 1.0 GBlops | Top-1 accuracy: 70.45 -> 73.29 | Improved |
| Use late-stage QAT finetuning rather than BN update alone (ablation) | NIPQ · CIFAR-100, MobileNet-v2, 1.0 GBlops | Top-1 accuracy: 72.99 -> 73.29 | Improved |
| Use a uniform PQN distribution instead of sampling the quantization-error distribution (ablation) | NIPQ · Quantization experiments | final quality | No clear effect |
Later papers that compared with this method
2 later accepted papers report NIPQ as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Efficient Quantization-Aware Adaptation for Visual Foundation Models (ECCV 2026 · better on 1 of 1 shared result)
- Temporal Dynamic Quantization for Diffusion Models (NeurIPS 2023 · better on 7 of 8 shared results)
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
- Leveraging Inter-Layer Dependency for Post -Training Quantization (NeurIPS 2022 · similarity 0.88)
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · similarity 0.88)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.87)
- SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning (ICLR 2025 · similarity 0.87)
- Quant Experts: Token-aware Adaptive Error Reconstruction with Mixture of Experts for Large Vision-Language Models Quantization (CVPR 2026 · similarity 0.87)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 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.
