Research map / Architectures and efficiency / Quantization
Adaptive Loss-Aware Quantization for Multi-Bit Networks
Check 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 |
|---|---|---|---|
| allocate adaptive bitwidth instead of uniform bitwidth (ablation) | ALQ on VGG · CIFAR10 | Top-1 accuracy: 91.8% -> 92.0% | Helped |
| allocate adaptive bitwidth instead of uniform bitwidth (ablation) | ALQ on ResNet18 · ILSVRC12 | Top-1 accuracy: 66.2% -> 68.9% | Helped |
| replace loss-aware binary-base optimization with an STE step using reconstruction forward (ablation) | ALQ on VGG · CIFAR10 | Top-1 accuracy: 92.0% -> 90.7% | Hurt |
| add extra epochs of STE with loss-aware after quantization (ablation) | ALQ (2.00/32) on ResNet18 · ILSVRC12 | Top-1 accuracy: 68.9% -> 70.0% | Helped |
| add extra epochs of STE with loss-aware after quantization (ablation) | ALQ (1.01/32) on ResNet18 · ILSVRC12 | Top-1 accuracy: 65.6% -> 67.7% | Helped |
| use adaptive loss-aware quantization rather than uniform bitwidth quantization (ablation) | ResNet18 · ILSVRC12, 2-bit weights | Top-1 validation accuracy | Helped |
| use ALQ's direct loss optimization rather than STE-based optimization (ablation) | uniform-bitwidth MBN optimizers · ImageNet validation | Top-1 validation accuracy | Mixed |
| retain a full-precision trace and use STE with loss-aware (ablation) | 2-bit ResNet18 optimizer · ILSVRC12 | Top-1 validation accuracy | Helped |
Closest papers by meaning
- Distribution-Aware Adaptive Multi-Bit Quantization (CVPR 2021 · similarity 0.94)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.92)
- Bit-Shrinking: Limiting Instantaneous Sharpness for Improving Post-Training Quantization (CVPR 2023 · similarity 0.91)
- Loss-aware Weight Quantization of Deep Networks (ICLR 2018 · similarity 0.91)
- ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization (NeurIPS 2025 · similarity 0.88)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 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.