Research map / Architectures and efficiency / Quantization
HAQ: Hardware-Aware Automated Quantization With Mixed Precision
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 |
|---|---|---|---|
| Use hardware-aware flexible mixed precision instead of uniform 4-bit quantization | MobileNet-V1 on BISMO edge · ImageNet, latency constraint | Acc.-1: 62.44 -> 67.40 | Helped |
| Use hardware-aware flexible mixed precision instead of uniform 6-bit weights and activations | MobileNet-V1 on BitFusion · ImageNet, latency constraint | Acc.-1: 70.46 -> 70.90 | Helped |
| Use hardware-aware flexible mixed precision instead of 6-bit weights and 4-bit activations | MobileNet-V1 on BitFusion · ImageNet, energy constraint | Acc.-1: 67.51 -> 70.37 | Helped |
| Search per-layer weight bitwidths rather than use uniform 2-bit weights | MobileNet-V1 · ImageNet, model size 1.09 MB | Acc.-1: 37.62 -> 57.14 | Helped |
| Search per-layer weight bitwidths rather than use uniform 3-bit weights | MobileNet-V2 · ImageNet, model size-constrained | Acc.-1: 68.00 -> 70.90 | Helped |
| Optimize the quantization policy separately for each hardware architecture | MobileNet-V1 · Inference latency across HW1, HW2, and HW3 | latency: 16.29 ms; 85.24 ms; 117.44 ms (HW1 policy) versus 19.95 ms; 64.29 ms; 108.64 ms (HW2 policy) versus 19.94 ms; 66.15 ms; | Mixed |
Closest papers by meaning
- HAWQ-V3: Dyadic Neural Network Quantization (ICML 2021 · similarity 0.91)
- JAQ: Joint Efficient Architecture Design and Low-Bit Quantization with Hardware-Software Co-Exploration (AAAI 2025 · similarity 0.91)
- BatchQuant: Quantized-for-all Architecture Search with Robust Quantizer (NeurIPS 2021 · similarity 0.89)
- Retraining-Free Model Quantization via One-Shot Weight-Coupling Learning (CVPR 2024 · similarity 0.88)
- Analog Foundation Models (NeurIPS 2025 · similarity 0.87)
- ScaleBITS: Scalable Bitwidth Search for Hardware-Aligned Mixed-Precision LLMs (NeurIPS 2026 · 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.