Research map / Architectures and efficiency / Quantization
Loss-aware Binarization of Deep Networks
Read the paperCheck 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 loss-aware binarization with diagonal curvature information instead of BWN's weight approximation | LAB · MNIST | classification error (%): 1.310 -> 1.180 | Helped |
| Use loss-aware binarization with diagonal curvature information instead of BinaryConnect | LAB · CIFAR-10 | classification error (%): 9.860 -> 10.500 | Hurt |
| Use loss-aware binarization with diagonal curvature information instead of BWN | LAB · SVHN | classification error (%): 2.535 -> 2.354 | Helped |
| Use loss-aware binarization with diagonal curvature information instead of BWN | LAB · War and Peace | cross-entropy: 1.313 -> 1.291 | Helped |
| Use loss-aware binarization with diagonal curvature information instead of BWN | LAB · Linux Kernel | cross-entropy: 1.307 -> 1.305 | Helped |
| Replace full LAB with LAB2, adding sign-binarized activations (ablation) | LAB · MNIST | classification error (%): 1.180 -> 1.380 | Hurt |
| Replace full LAB with LAB2, adding sign-binarized activations (ablation) | LAB · CIFAR-10 | classification error (%): 10.500 -> 12.280 | Hurt |
| Replace full LAB with LAB2, adding sign-binarized activations (ablation) | LAB · SVHN | classification error (%): 2.354 -> 3.362 | Hurt |
| Replace full LAB with LAB2, adding sign-binarized activations (ablation) | LAB · War and Peace | cross-entropy: 1.291 -> 1.376 | Hurt |
| Replace full LAB with LAB2, adding sign-binarized activations (ablation) | LAB · Linux Kernel | cross-entropy: 1.305 -> 1.409 | Hurt |
| Increase CNN filter count from K = 16 to K = 128 (ablation) | LAB · SVHN | classification error (%): 3.050 -> 2.200 | Helped |
| Increase LSTM unrolled time steps from TS=10 to TS=100 (ablation) | BinaryConnect · War and Peace | cross-entropy: 1.528 -> 2.942 | Hurt |
Closest papers by meaning
- Loss-aware Weight Quantization of Deep Networks (ICLR 2018 · similarity 0.90)
- Forward and Backward Information Retention for Accurate Binary Neural Networks (CVPR 2020 · similarity 0.88)
- Regularizing Activation Distribution for Training Binarized Deep Networks (CVPR 2019 · similarity 0.88)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.87)
- Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms (NeurIPS 2024 · similarity 0.86)
- Layerwise Progressive Freezing: A Training Scaffold for Depth-Scalable Binary Networks (NeurIPS 2026 · similarity 0.84)
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.