Research map / Optimization / Optimizers
Demystifying and Generalizing BinaryConnect
Tim Dockhorn, Yaoliang Yu, Eyyub Sari, Mahdi Zolnouri, Vahid Partovi Nia
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 ProxConnect instead of BinaryConnect | ResNet20 ternary · CIFAR-10 end-to-end | test accuracy: 27.10 -> 84.09 | Improved |
| use ProxConnect instead of BinaryConnect | ResNet56 ternary · CIFAR-10 end-to-end | test accuracy: 30.31 -> 84.36 | Improved |
| use ProxConnect instead of ProxQuant | ResNet20 binary · CIFAR-10 end-to-end | test accuracy: 81.59 -> 89.92 | Improved |
| use ProxConnect instead of BinaryConnect | ResNet20 binary · CIFAR-10 end-to-end | test accuracy: 87.51 -> 89.92 | Improved |
| use ProxConnect instead of BinaryConnect | ResNet56 quaternary · CIFAR-10 fine-tuning | test accuracy: 93.20 -> 93.42 | Improved |
| use ProxConnect instead of BinaryConnect | ResNet20 binary · CIFAR-10 fine-tuning | test accuracy: 90.31 -> 90.31 | No clear effect |
| use ProxConnect instead of BinaryConnect | ResNet18 ternary-free binary quantization · ImageNet fine-tuning | test accuracy: 65.84 -> 65.70 | Hurt |
| use ProxConnect instead of BinaryConnect | ResNet18 · ImageNet end-to-end | test accuracy: 63.79 -> 63.89 | No clear effect |
| train from random init instead of fine-tuning pretrained | ProxQuant · CIFAR-10 ResNet20 binary | test accuracy: 89.94 -> 81.59 | Hurt |
| train from random init instead of fine-tuning pretrained | BinaryConnect · CIFAR-10 ResNet20 ternary | test accuracy: 74.95 -> 27.10 | Hurt |
| train from random init instead of fine-tuning pretrained | ProxConnect · CIFAR-10 ResNet20 ternary | test accuracy: 91.37 -> 84.09 | Hurt |
| use BinaryConnect for ternary quantization instead of ProxQuant | ResNet20 · CIFAR-10 fine-tuning | test accuracy: 91.46 -> 74.95 | Hurt |
Closest papers by meaning
- Understanding Neural Network Binarization with Forward and Backward Proximal Quantizers (NeurIPS 2023 · similarity 0.91)
- Proximal Mean-Field for Neural Network Quantization (ICCV 2019 · similarity 0.87)
- ProxQuant: Quantized Neural Networks via Proximal Operators (ICLR 2019 · similarity 0.87)
- Efficient and Generalizable Mixed-Precision Quantization via Topological Entropy (NeurIPS 2025 · similarity 0.86)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.86)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.85)
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.
